JPH08202822A - Character cutting device and character cutting method - Google Patents

Abstract

(57) [Abstract] [Object] The present invention relates to a character clipping device and a character clipping method, and an object thereof is to reliably extract a character by extracting a change point even if a continuation line cannot be extracted. And [Constitution] In the device A shown in FIG. A, a change point extraction unit 10 for extracting a change point of inclination by a pattern contour search, and a separation point determination unit 1 for determining a character separation point based on the extracted change point.
1. Character separation unit 1 that separates characters based on character separation points
2 and so on. In the device B of FIG. B, when two or more continuation lines are extracted, the continuation character extracting unit 23 that determines that the correction line is not the continuation line but the correction line written on the character and extracts the correction line,
Based on the position information of the continuation line obtained by the continuation line extraction unit 4, a blank line between characters is searched for with a polygonal line, and when the polygonal line reaches the bottom of the pattern, it is determined that the pattern can be separated. Part 5 and the like are provided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character slicing device and a character slicing method used in an OCR handwritten character recognition device or the like.

In recent years, as a peripheral device for inputting handwritten characters,
The demand for OCRs with handwriting recognition devices is increasing rapidly. In order to realize a high recognition rate of individual characters, the character segmentation process, which is the previous stage of character recognition, is important in terms of its accuracy.

Up to now, if the characters such as a form were separated and written one by one, the characters could be recognized at a considerable rate, but characters that are in contact with each other or continuous characters are included. And the recognition rate drops significantly. Under such circumstances, a method for accurately recognizing the continuation characters one by one is demanded in the handwritten character recognition device.
It is required to improve the accuracy of the character cutting process in the character cutting device.

The present invention is a technique applicable not only to handwritten character recognition but also to character cutting processing in a broad sense such as recognition of printed characters and character cutting in drawing recognition.

[0005]

18 to 25 are views showing a conventional example. In FIGS. 18 to 25, 1 is a character slicing device, 2 is a connected pattern extraction unit, 3 is a horizontally long pattern extraction unit, and 4 is a continuation. Line extraction unit, 5 is a character separation line determination unit, 6 is a zero determination unit, 7
Indicates a character separation part.

§1: Description of Configuration of Character Extraction Device--See FIG. 18 FIG. 18 is a configuration diagram of a conventional character extraction device. Conventionally,
When character recognition is performed in the character recognition device, it is necessary to perform a process of cutting out handwritten characters as a process before the character recognition. Therefore, the character slicing device was used.

As shown in the figure, the character slicing device 1 is provided with a concatenated pattern extraction unit 2, a horizontally long pattern extraction unit 3, a continuous line extraction unit 4, a character separation line determination unit 5, a zero determination unit 6, and a character separation unit 7. There is. The function of each part is as follows.

(1): The connection pattern extraction unit 2 extracts a connection pattern from an input pattern. That is, when performing character cutting, first, in order to stably pick up each character pattern without depending on the relative relationship of the positions where the characters are arranged, the connected pattern extraction unit 2 may, for example, connect 8 characters. The patterns connected by are extracted by the labeling process.

(2): The horizontal pattern extracting unit 3 calculates the vertical / horizontal ratio of the circumscribed rectangle for each concatenated pattern obtained by the labeling process as a candidate for the continuous character, and the horizontal ratio of a certain level or more. The pattern that has is extracted.

(3): The continuation line extraction unit 4 finds out whether or not the extracted connection pattern is a continuation character by finding a continuation line of a portion where characters are connected horizontally or diagonally. Judgment is made and a horizontal continuation line or a diagonal continuation line is extracted.

(4): The character separation line determination unit 5 obtains the number of characters and the area where each character is present by using contour search based on the extracted continuation line, and determines the space between characters. A character separation line such as a vertical separation line or a diagonal separation line is drawn on the.

(5): The zero determining unit 6 determines, for each character, for each character area separated by the character separation line.
It is to determine whether the number is zero (0). (6): The character separation unit 7 erases unnecessary continuation lines appearing on the left and right for the character determined to be the number zero (0), and the characters other than zero are separated by the character It is separated by a separation line.

§2: Outline of character cutout method ... See FIG. 18. The character cutout method by the character cutout device configured as described above is as follows. When the input pattern is input, the connection pattern extraction unit 2 extracts a connection pattern in which pixels are connected from the input pattern based on the connection information of the input pattern.

Next, the horizontally long pattern extraction unit 3 extracts only horizontally long patterns from the extracted connected patterns as candidates for a continuous character in which a plurality of characters are connected. Then, the continuation line extraction unit 4 determines whether or not the extracted connection pattern is a continuation character by finding a continuation line of a portion where characters are connected horizontally or diagonally, and horizontal, Alternatively, the diagonal continuation line is extracted.

Thereafter, the character separation line determining unit 5 uses the contour search of the pattern based on the extracted continuation line to obtain the number of characters and the area where each character exists, and the area between the characters is determined. Character separation lines such as vertical separation lines or diagonal separation lines are determined.

When the character separation line is determined, the zero determination unit 6 determines whether or not the character area of each character separated by the character separation line is a numeral zero (0) for each character. judge.

When this process is completed, the character separation unit 7 erases unnecessary continuation lines appearing on the left and right of the character determined to be the numeral zero (0), and the character other than the numeral zero. Are separated by the character separation line. In this way, the characters are cut out. Hereinafter, detailed processing of each unit will be described.

§3: Detailed processing description of the connection pattern extraction unit, the horizontally long pattern extraction unit, and the continuous line extraction unit ... FIG.
Refer to FIG. 9. FIG. 19 is a process explanatory diagram of the connection pattern extraction unit, the horizontally long pattern extraction unit, and the continuous line extraction unit.

(1): Usually, when a continuous character is extracted and separated from a freely written character string, for example, a numerical character string, regarding the character, a single character and a continuous character are mixed,
There are no particular restrictions on the size of characters or the number of characters that are written in succession. Therefore, extreme inclination correction is performed on the input character pattern, noise is removed, and a binary image after preprocessing such as fill-in of faintness is targeted for character extraction.

Paying attention to the fact that when characters are written in succession, the characters are often connected by horizontal lines.
By extracting the horizontal continuation line (hereinafter referred to as "horizontal continuation line") or the diagonal continuation line (hereinafter referred to as "diagonal continuation line"), the continuation character is determined and the characters are separated. I am trying.

(2): When character extraction is performed, first, in order to stably pick up each character pattern without depending on the relative relationship of the positions where characters are arranged, the connection pattern extraction unit 2 , For example, patterns connected by 8 connections are extracted by the labeling process.

Specifically, the character pattern is scanned with an 8-connected window pattern, and the same label is given to the pixels which can be scanned with the window based on a predetermined logic.
Hereinafter, these labels are discriminated and the subsequent characters are separated.

Since the size of the concatenated pattern obtained by the labeling process is required later, the coordinates of the corners of the rectangle obtained by approximating the concatenated pattern with a rectangle are calculated by the labeling process. To do so.

Regarding the labeling process, for example, "Basic Techniques of Image Processing (Introduction to Techniques)", (Image Proces
sing on Personal Computer), Part 1 Basics of Image Processing, Chapter 3, Basic Algorithm of Image Processing, 2) Connected Component Processing, Labeling, Junichi Hasegawa, Yamato Komizu, Akira Nakayama, Shigeki Yokonaka, Technical Review , August 10, 1986 ”.

(3): In the next horizontally long pattern extraction unit 3, the vertical and horizontal ratios of the circumscribing rectangle are calculated for each concatenated pattern obtained by the labeling process as a candidate for the continuous character, and a certain value or more is calculated. The pattern with the horizontal ratio of is extracted.

Further, in this processing, for example, the average size of the extracted character region is calculated by approximating the concatenated pattern to a rectangle, and is compared with the calculated average size to extract the candidate pattern of the subsequent character.

(4): In the continuous line extracting unit 4, the horizontal continuous line,
The diagonal continuous line is extracted. In the horizontal continuous line extraction processing, first, (pattern area) / (area of circumscribed rectangle) is calculated, and the length of the straight line to be extracted is determined according to the value.

When the value is large, a long straight line is extracted as shown in FIG. 19A, and when the value is small, a short straight line is extracted as shown in FIG. 19B. The reason why the length of the extracted straight line is not constant will be described below.

(5): As shown in FIGS. 19A and 19B, when (pattern area) / (circumscribed rectangle area) = large, a short straight line as shown in FIG. 19A Is extracted, a large number of straight lines are also extracted in the character portion. Therefore, in order to extract a straight line only in the continuous portion, it is necessary to extract a relatively long straight line as shown in FIG.

On the other hand, the above (pattern area) /
When (area of circumscribing rectangle) = small, there is a case where a straight line is not extracted even if an attempt is made to extract a long straight line as shown in FIG. 19B. Therefore, as shown in FIG. 19B, This is because a method for obtaining a long continuous line is required by extracting short straight lines and integrating them.

(6): Next, horizontal projection is performed to extract straight lines. At this time, a so-called adjacent projection method is used in which the projection value of the surrounding row is added to the projection value to obtain the projection value of the row. By using this adjacent projection method, it is possible to detect a continuous line that is diagonally connected.

FIG. 19C is a diagram for explaining the projection range when performing the adjacent projection. In the projection range when performing the adjacent projection, the character pattern is vertically divided according to the length of the extracted straight line, and the adjacent projection is performed within the divided range.

When the projection value is a certain value or more, it is recognized that there is a straight line, and the range in which the extracted straight line exists is approximated to a rectangle as shown in FIG. 19D. Then, a rectangular straight line is formed.

After that, as shown in FIG. 19D, the contact or neighboring rectangular straight lines are integrated to extract a long straight line, and the longest straight line among them is set as a horizontal continuation line. This horizontal continuation line also uses a rectangular straight line subjected to rectangular approximation.

When the horizontal continuation line is extracted as described above, the continuation character is discriminated and the characters are separated. However, when the continuation line is oblique, the horizontal continuation line extraction processing cannot extract the horizontal continuation line. There is. In such a case, in addition to the horizontal continuation line extraction processing, the diagonal continuation line extraction processing is performed by performing adjacent projection in an oblique direction.

§4: Detailed description of diagonal continuation line extraction processing by the continuation line extraction unit ... See FIGS. 20 to 22. FIG. 20 is a process explanatory diagram (No. 1) of the continuation line extraction unit.
FIG. A is an explanatory diagram of a method for calculating a diagonal projection value, and FIG. B is an explanatory diagram for an oblique adjacent projection method. FIG. 21 is a process explanatory diagram (No. 2) of the continuous line extraction unit, FIG. A is an explanatory diagram of an original image, B is an explanatory diagram of straight line extraction (inclination of 45 degrees), and C is an oblique projection value calculation. FIG. FIG. 22 is a process explanatory diagram (part 3) of the continuous line extraction unit, in which FIG. A shows pattern division, FIG. B shows an extracted rectangular straight line, and FIG. C shows an integratedly extracted long straight line.

As described above, when the horizontal continuation line extraction processing of the continuation line extraction unit 4 fails to extract the horizontal continuation line, the diagonal continuation projection method described below is performed to obtain an oblique continuation line. Extract the line.

(1): Adjacent projection method in diagonal direction. See FIGS. 20 and 21. The adjacent projection method in diagonal direction projects in a certain diagonal direction, and projects the diagonal line of interest. In this method, the projection value of the diagonal line is the result of adding the projection values of its surroundings to the value. FIG. 20A shows a method of calculating a projection value when an n × n area is obliquely projected at an angle of 45 degrees.

In this example, the projection value in the vertical direction is P1 (i).
(I = 1 to n), and the projection value in the horizontal direction is P2 (j) (j = 1
˜n) and P1 (n) = P2 (1). In addition, I (x, y) is a pixel value in the formula described below.
In this case, the formulas for calculating the projection values P1 (i) and P2 (j) are as follows.

[0040]

[Equation 1]

As shown in FIG. 20B, the calculation of the oblique projection value can be expressed by the following equation, assuming that the i-th projection value is P (i). . Note that z represents the number of projection values in the surroundings to be added.

First, in the case of the upper left area of the figure, P (i)
= P1 (i-z) + p1 (i-z + 1) + ... + p1
(I) + ... + p1 (i + z). In the case of the upper right area, P (j) = p2 (jz) + p2 (j-
z + 1) + ... + p2 (j) + ... + p2 (j +
z).

When i = z, P (i) = p1
(I-z) + p1 (i-z + 1) + ... + p1 (i)
... + p1 (z). Note that in the case of FIG. 20B, the diagonal adjacent projection value is P (i) = p (i-1) + p
(I) + p (i + 1).

A specific example will be described with reference to FIG. For example, assume that there is an original image as shown in FIG.
In this figure, each pixel is shown by a square mark. In the case of performing the straight line extraction, as shown in FIG. 21B, when the inclination is 45 degrees in the XY coordinates, the process advances by 1 in the X direction and then by 1 in the Y direction. Similarly, when the angle is other than 45 degrees, the integer ratio of X and Y is calculated, and the projection value is calculated according to the ratio.

For example, if z = 1 and the threshold value = 10 in the above equation, in the case of FIG. 21A, the 10th line is extracted as a straight line. Further, in FIG. 21C, the ratio of the increase number in the X direction and the increase number in the Y direction is 1: 2.

(2): Explanation of the process from the calculation of the diagonal adjacent projection values to the detection of the diagonal continuation line ... See FIG. 22. In the diagonal adjacent projection value calculation process, as shown in FIG.
Divide the pattern into horizontal multiples, and within each division range,
Perform diagonal adjacent projection. In this case, it is assumed that a straight line exists in a row in which the ratio between the adjacent projection value and the division length is a certain fixed value or more.

In the oblique straight line extraction processing, as shown in FIG. 22B, a rectangular straight line is extracted by approximating the range in which the extracted straight line exists to a rectangle. Then, diagonal straight line integration processing is performed, and as shown in FIG. 22C, a long straight line is extracted by integrating contact or neighboring rectangular straight lines,
The longest straight line among the integrated straight lines is the diagonal continuation line. Then, the diagonal continuation line is approximated to a rectangle and output.

§5: Process Description of Character Separation Determining Unit ... See FIG. 23. FIG. 23 is a process explanatory diagram of the character separation line determining unit. FIG. A is a pattern inclination calculation, B is a pattern inclination calculation, and FIG. FIG. C is an explanatory diagram of determining character separation lines.

In the character separation line determination unit 5, the continuation line extraction unit 4
The character separation line is determined based on the continuation line extracted in. In this process, when a horizontal continuation line is extracted, the lower side of the horizontal continuation line approximated to the rectangle is traced from the end, and the contour search is performed with the intersection with the pattern as the starting point of the contour search.

Then, when the lower side is crossed again, the search is ended, and the lower side is traced until the pattern crosses the pattern again, and the same processing is repeated. The number of times the search is finally performed is defined as the number of characters, and the area from the start point to the end point of the search is the area where one character exists. A portion between the search end point and the start point and having a vertical line density of 1 is defined as a vertical separation line for separating characters.

If there is no portion having a line density of 1 in the vertical direction, character separation is performed by diagonal lines. In this case, the inclination of the separation line uses the inclination of the pattern. The inclination of the pattern is calculated, for example, using the formula described below. For details of this process, refer to the following document, for example.

"F. Kimura, M. Shridhar and Z. Chen" Imp
rovement of Lexicon Directed Algorithm Recognition
of Unconstrained Handwritten Words "Proceeding of
Second International Conference on Document Analy
sis and Recognition, Tsukuba Science City, Japan 19
93, IEEE Computer Society Press, P.18-P.22 ”.

As described above, the line density is obtained between the character regions according to the inclination, and the separation line (oblique separation line) having the inclination is drawn at the line density 1. When the inclination of the pattern is calculated, for example, as shown in FIGS. 23A and 23B, it is assumed that the directions of the adjacent pixels n1, n2, and n3 are in the arrow directions shown. Such an adjacent pixel n1,
For n2 and n3, the total number of n1 is N1 and the total number of n2 is N
When the total number of 2 and n3 is N3 and the inclination angle of the pattern is θ, tan θ = (N1 + N2 + N3) / (N3-N
The formula of 1) is established.

That is, as shown in FIG. 23C,
The angle θ of the oblique separating line with respect to the horizontal axis is tan θ = (N
1 + N2 + N3) / (N3-N1). §6: Detailed process description of zero determination unit ... See FIG. 24. FIG. 24 is a process explanatory diagram of the zero determination unit. FIG. A shows linear density calculation processing, and FIG. B shows linear density calculation processing.

The zero decision unit 6 has a continuation line (horizontal continuation line and diagonal continuation line) and a separation line (vertical separation line, oblique separation line).
As shown in FIG. 24, the line density is calculated in a radial direction in a plurality of directions from the blank portion surrounded by the continuation line and the pattern, and if the line density is 1 in all directions, as shown in FIG. , It is determined that the loop structure is zero.

§7: Detailed Process Description of Character Separation Unit--See FIG. 25 FIG. 25 is a process explanatory diagram of the character separation unit. Character separation part 7
Then, for the character that is determined to be zero by the zero determination unit 6, unnecessary continuation lines appearing on the left and right are deleted,
Other characters are separated by a character separation line.

In this case, the erasing range is from the separation line to the portion where the slope of the line changes abruptly. As shown in FIG. 25A, when there is almost no change in inclination within a certain range,
The continuation line is not erased. That is, the portion a is a portion which is not erased because there is no abrupt change in the inclination, and the portion b is a portion which is erased to a portion where the inclination is abruptly changed.

If there is a loop inside the pattern determined to be zero as shown in FIG. 25B, in order to reduce the adverse effects such as erroneous recognition in the subsequent recognition processing, It also erases the line. For characters determined to be non-zero, the separating line separates them from other characters, and the horizontal continuation line is not erased.

[0059]

SUMMARY OF THE INVENTION The above-mentioned conventional device has the following problems. (1): It was possible to perform character recognition with a considerably high recognition rate for characters written separately in documents such as forms. However, when a contact character or a continuous character is entered, the character cannot be separated from each other in many cases, and the recognition rate of the character decreases.

(2): In the case of a continuous character of zeros in a continuous line, it is possible to accurately perform character separation processing. However, if there was a continuation line and the number was not zero, it was often not possible to find the exact line between the characters. In such a case, the characters are separated from each other by performing the forced separation at a predetermined position.

Therefore, due to the forced separation, an extra "whisker" remains, and the recognition process becomes difficult, and other adverse effects occur. (3): In the conventional apparatus, when extracting a diagonal continuation line, projection processing is performed in a direction having a certain angle to extract the continuation line. In this process, if the initial angle is set incorrectly, the subsequent line cannot be extracted accurately.

(4): When separating consecutive characters, since it is checked whether two characters can be separated by a straight line, it cannot be dealt with if the space between the characters becomes a complicated shape. Moreover, the correction line of a character may be mistakenly extracted as a continuation line.

The present invention solves such a conventional problem, and enables highly accurate recognition processing by always accurately separating characters from each other even in the case of a character that does not include the number zero. The purpose is to enable.

Further, according to the present invention, it is possible to improve the extraction accuracy of the continuation line by extracting the correction line of the character and to realize the extraction processing of the change point with high accuracy to improve the accuracy of character separation. To aim.

[0065]

FIG. 1 is a diagram for explaining the principle of the present invention. FIG. 1A shows a character slicing device a and FIG. B shows a character slicing device b. In order to achieve the above-mentioned object, the present invention has a character cutting device configured as follows.

(1): In the character slicing device B shown in FIG. 1A, the connection pattern extracting unit 2 for extracting the connection pattern from the input pattern and the connection pattern extracting unit on the basis of the connection information of the input pattern. Among the extracted concatenated patterns, a horizontal pattern extracting unit 3 that extracts only a horizontally long horizontal pattern as a candidate for a continuous character in which a plurality of characters are connected, and characters and characters are extracted from the horizontal pattern extracted by the horizontal pattern extracting unit. It is determined whether or not the extracted connection pattern is a continuation character by finding the continuation line that is connected, and the continuation line extraction unit 4 for extracting the continuation line of the character and the continuation line extraction unit If the extraction is not possible, the pattern contour search is performed to extract a change point of the inclination of the pattern contour, and the change point extraction unit 10 that makes the change point a character and character separation point candidate. Based on the change points extracted by the change point extraction unit, the size of the character after separation when the separation point candidates are separated is obtained, and the character separation point is determined by comparing the character sizes. A separation point determination unit 11 and a character separation unit 12 that separates characters at the character separation points determined by the separation point determination unit are provided.

(2): In the character slicing device B shown in FIG. 1B, the connection pattern extracting unit 2 for extracting the connection pattern from the input pattern and the connection pattern extracting unit on the basis of the connection information of the input pattern. Of the extracted concatenated patterns, horizontal pattern extraction processing that extracts only horizontally long horizontal patterns as candidates for continuous characters in which multiple characters are connected, and continuous lines that connect characters from the horizontal patterns are extracted. Then, by detecting the continuous character extracting unit 23 that performs the correction line extracting process for extracting the correction line written on the character, and the continuous line connecting the characters from the horizontally long pattern extracted by the continuous character extracting unit, When the continuation line is extracted by the continuation line extraction unit 4 that determines whether the extracted connection pattern is a continuation character and extracts the continuation line of the character. , The character separation line determination unit 5 that determines the character separation line based on the position information of the continuation line, and the character area of each character divided by the character separation line determined by the character separation line determination unit , A zero determination unit 6 for determining whether it is a numeral zero, a character separation unit 7 for leaving a character area and erasing an unnecessary continuation portion for a character determined to be a numeral zero, and a continuation line extraction unit When the continuation line cannot be extracted in step 4 and when the zero determining unit does not determine that the number is zero, a pattern contour search is performed to extract a change point of the inclination of the pattern contour, and the change point is determined. Based on the change point extraction unit 10 as a character and character separation point candidate, and the change point extracted by the change point extraction unit, the size of the separated character when separated by each separation point candidate is obtained, By comparing the size of the characters, A separation point determining unit 11 to a constant, a character separated by the separation point of the character determined in the separation point determining unit 11, provided with the character segmentation unit 12 to erase the unnecessary continuation portion leaving the character region.

[0068]

The operation of the present invention based on the above construction will be described with reference to FIG. (1): In the character slicing device B, first, the connection pattern extraction unit 2 extracts a connection pattern in which pixels are connected from the input pattern based on the connection information of the input pattern.

Next, the horizontally long pattern extraction unit 3 extracts only horizontally long horizontally long patterns from the extracted connected patterns as candidates for a continuous character in which a plurality of characters are connected. Then, the continuation line extraction unit 4 determines whether or not the extracted connection pattern is a continuation character by finding a continuation line connecting characters to extract a horizontal or diagonal continuation line. .

After that, when the continuation line can be extracted, another processing unit (not shown) determines a character separation point based on the continuation line to separate the characters. However, when the continuation line cannot be extracted, the change point extraction unit 10 searches for the contour of the pattern to extract the change point of the inclination of the pattern contour,
The change point is defined as a character-to-character separation point candidate.

Thereafter, the separation point determination unit 11 obtains the size of the character after the separation when the separation point candidates are separated on the basis of the change point extracted by the change point extraction unit 10, and the size of the character. The character separation point is determined by comparing the character sizes. Next, the character separation unit 12 separates the characters at the character separation points determined by the separation point determination unit 11.

(2): In the character slicing device B, the connection pattern extraction unit 2 first extracts a connection pattern in which pixels are connected from the input pattern based on the connection information of the input pattern.

Next, the continuous character extracting unit 23 extracts only horizontally long horizontal patterns as continuous character candidates in which a plurality of characters are connected, among the connected patterns extracted by the connected pattern extracting unit 2. At this time, the continuous character extracting unit 23 extracts a correction line written on the character by extracting a continuous line connecting the characters from the horizontal pattern.

After that, the continuation line extraction unit 4 finds a continuation line connecting characters from the horizontally long pattern extracted by the continuation character extraction unit 23 to determine whether the extracted connection pattern is a continuation character. Then, the continuation line of the character is extracted.

Then, when the continuation line is extracted by the continuation line extraction unit 4, the character separation line determination unit 5 determines the character separation line based on the position information of the continuation line. Then, the zero determination unit 6
Determines whether or not the number is zero for each character in the character area for each character divided by the character separation line determined by the character separation line determination unit 5. As a result, when it is determined that the number is zero, the character separating unit 7 leaves a character area and erases an unnecessary continuous portion of the character that is determined to be zero.

However, when the continuation line extraction unit 4 cannot extract the continuation line and when the zero determination unit 6 does not determine that the number is zero, the change point extraction unit 10 searches for the contour of the pattern. Then, the change point of the inclination of the pattern contour is extracted, and the change point is used as a character and character separation point candidate.

Next, the separation point determination unit 11 is connected to the change point extraction unit 1
Based on the change points extracted by 0, the size of the character after the separation when the separation point candidates are separated is obtained, and the character separation point is determined by comparing the sizes of the characters.

Then, the character separation unit 12 separates the characters at the character separation points determined by the separation point determination unit 11, leaving a character area and erasing unnecessary continuation portions. Characters are cut out by the above processing.

(3): The character slicing device B shown in FIG. A has the following effects. a) In the processing of the change point extraction unit 10, after determining a tentative search start reference line based on the vertical length of the pattern of interest, the average size of other patterns, etc., a line is formed around the tentative search start reference line. The formal search start reference line is obtained by searching the portion of density 1, the search start point is determined from the search start reference line, and the contour search of the pattern is performed.

B) In the processing of the change point extraction section 10, the line density 1 is set as the search start point when the contour search of the pattern is performed.
There are two points, the upper end and the lower end of the pattern of the part. c) In the process of the change point extraction unit 10, after the search start point for starting the contour search of the pattern is determined, the contour search is performed in the left and right directions of the pattern from the search start point to obtain the change point of the inclination of the pattern. To extract.

D) In the processing of the change point extraction unit 10, when the contour search of the pattern is performed to extract the change point of the inclination of the pattern contour, the search start point on the pattern is connected to a certain point on the contour. The change point is extracted by calculating the overlap between the straight line and the detected contour, and determining whether or not the search start point to a certain point on the contour is a straight line based on the calculated degree of overlap. To do.

E) In the process of the change point extraction section 10, when the contour of the pattern is searched to extract the change point of the inclination of the contour of the pattern, the search start point on the pattern is connected to a certain point on the contour. The overlap between the straight line and the contour is calculated, and the change point of the inclination of the contour is extracted from the change amount of the overlap.

F) In the processing of the separation point determination unit 11, if the distance between the change points of the upper and lower two points of the pattern is less than a certain threshold value, the two points are selected as the change points, When the distance exceeds a certain threshold value, one point closer to the search start point is selected as the change point.

G) In the process of f), the threshold value of the distance between the upper and lower two points is extracted by the adjacent projection method in which the projection value of the target row or column is the result of adding the projection values of the surroundings. Set based on the average thickness of the line segments.

H) In the processing of f), the threshold value of the distance between the upper and lower points is set with reference to the height of the pattern of interest. i) In the processing of the character separation unit 12, the separation point determination unit 11
When a character is separated at the character separation point determined in step 1, the line connecting the separation points and the pixels on both sides thereof are erased.

J) In the processing of the separation point determination unit 11, it is assumed that the separation points are separated by respective separation point candidates based on the change points extracted by the change point extraction unit 10, and labeling, contour search,
The separation point is determined by obtaining the size of the character after separation by a projection method or the like.

K) In the processing of the separation point determination unit 11, it is assumed that the respective separation point candidates are separated based on the change points extracted by the change point extraction unit 10, and the size of the character after the separation is obtained, Separation points are determined so that the sizes after separation are equal.

(4): The character slicing device B shown in FIG. a) In the processing of the continuous character extracting unit 23, when extracting the continuous characters from the continuous character candidates, if two continuous lines are extracted by performing the projection process in the horizontal direction, the two continuous lines are determined as correction lines. However, the character is treated as a correction character.

B) In the processing of the continuation line extraction unit 4, when extracting the continuation line connecting characters from the horizontally long pattern extracted in the processing of the continuation character extraction unit 23, the continuation line is extracted for a plurality of slopes. Perform extraction processing.

C) In the processing of the character separation line determination section 5,
When the continuation line is extracted by the continuation line extraction unit 4 and the character separation line is determined based on the position information of the continuation line, whether or not the character can be separated is determined by whether or not the two characters can be separated by the polygonal line.

D) In the process of the change point extraction unit 10, when extracting change points as points where characters touch each other, first find a point on the contour line having a line density of 1 in the vertical direction, and then change points. A plurality of search start points are set on the contour line, and the plurality of change point candidates are calculated from each point.

E) In the process of d), when a plurality of search start points for changing points are set on the contour line, the search range for the pattern is limited to find the search start points. f) In the process of d), when calculating a plurality of change point candidates, the change range candidates are calculated by limiting the search range for the pattern.

G) In the process of e), when the search range of the search start point is limited, a constant multiple of the height of the pattern is set as the search range. h) When the search range of the search start point is limited in the process of e), the search range is set to a constant multiple of the average character size of the character string to which the character belongs.

I) In the process of the above-mentioned f), when the search range of the changing point is limited, a constant multiple of the height of the pattern is set as the search range. j) In the process of f), when limiting the search range of the change point, the search range is set to a constant multiple of the average character size of the character string to which the character belongs.

K) In the processing of the character separation units 7 and 12,
The whiskers of the characters that occur when the continuous characters are separated are removed, and at the same time, the whiskers included in the characters are removed from the position information of the upper contour line and the continuation line of the pattern.

(5): As described above, even in the case of a character that does not include the number zero, high-accuracy recognition processing can be performed by always allowing accurate character separation. be able to.

Further, by extracting the correction line of the character, it is possible to improve the extraction accuracy of the continuation line, realize the change point extraction processing with high accuracy, and improve the character separation accuracy.

[0098]

Embodiments of the present invention will be described below with reference to the drawings. 2 to 17 are diagrams showing an embodiment of the present invention. In FIGS. 2 to 17, the same parts as those in FIGS. 1 and 18 to 25 are designated by the same reference numerals. Further, 14 is a search start reference line determination unit, 15 is a search start point determination unit, 16 is a contour search unit, 17 is a change point extraction processing unit, 19 is a separation point candidate extraction unit, 20 is a size comparison unit, and 21 is The separation line determination part is shown.

(Explanation of Embodiment 1) §1: Description of basic configuration of character slicing device--see FIG. 2 FIG. 2 is a basic configuration diagram of the character slicing device of the first embodiment.
As illustrated, the character slicing device 1 includes a connection pattern extraction unit 2, a horizontally long pattern extraction unit 3, a continuous line extraction unit 4, a character separation line determination unit 5, a zero determination unit 6, a character separation unit 7, and a change point extraction unit. 10, a separation point determination unit 11, and a character separation unit 12.

Of the above-mentioned units, the configuration other than the change point extraction unit 10, the separation point determination unit 11, and the character separation unit 12 is the same as that of the conventional example. The function of each part is as follows. (1): The connection pattern extraction unit 2 extracts a connection pattern from the input pattern. That is, when performing character segmentation, first, in order to stably pick up each character pattern without depending on the relative relationship of the positions where the characters are arranged, the connected pattern extraction unit 2
For example, a pattern connected by eight connections is extracted by the labeling process.

(2): The horizontally long pattern extraction unit 3 calculates the vertical / horizontal ratio of the circumscribed rectangle for each concatenated pattern obtained by the labeling process as a candidate for the continuous character, and the horizontal / horizontal ratio above a certain level. The pattern that has is extracted.

(3): The continuation line extraction unit 4 finds whether or not the extracted connection pattern is a continuation character by finding a continuation line of a portion where characters are connected horizontally or diagonally. Judgment is made and a horizontal continuation line or a diagonal continuation line is extracted.

(4): The character separation line determination unit 5 obtains the number of characters and the area in which each character exists by using contour search based on the extracted continuation line, and determines the space between characters. A character separation line such as a vertical separation line or a diagonal separation line is drawn on the.

The character separation line determination unit 5 determines the character separation line only when the continuation line extraction unit 4 extracts a horizontal continuation line or an oblique continuation line (when there is a continuation line). It is a thing.

(5): The zero determining unit 6 determines, for each character, for each character area separated by the character separation line.
It is to determine whether the number is zero (0). (6): The character separation unit 7 erases unnecessary continuation lines and whiskers appearing on the left and right of the character determined to be zero (0) of the numeral, and separates by the character separation line. Is used to separate characters.

(7): If the continuation line extraction unit 4 cannot extract a continuation line (horizontal continuation line or diagonal continuation line) by the processing of the continuation line extraction unit 4 (when there is no continuation line), the processing is performed. The change point of the inclination is extracted from the target pattern.

(8): The separation point determination unit 11 determines the character separation point based on the change points extracted by the change point extraction unit 10. (9): The character separation unit 12 performs character separation based on the character separation points determined by the separation point determination unit 11.

§2: Description of Configuration of Character Extraction Device of First Embodiment--See FIG. 3 FIG. 3 is a configuration diagram of the character extraction device of the first embodiment. As shown in the figure, the character slicing device includes a connected pattern extraction unit 2, a horizontally long pattern extraction unit 3, a continuous line extraction unit 4, a character separation line determination unit 5, a zero determination unit 6, a character separation unit 7, and a change point extraction unit 10. , A separation point determination unit 11 and a character separation unit 12.

The change point extraction unit 10 is composed of a search start reference line determination unit 14, a search start point determination unit 15, a contour search unit 16, and a change point extraction processing unit 17, and the separation point determination unit 11 is , Separation point candidate extraction unit 19, size comparison unit 2
0, separating line determining unit 21. Functions and the like of the above-mentioned respective parts are as follows.

(1): When the continuation line is not extracted, the search start reference line determination unit 14 determines the search start reference line for searching for a change point such as a continuation character or a touch character. It is a thing.

(2): The search start point determination unit 15 determines the search start point based on the search start reference line determined by the search start reference line determination unit 14. (3): The contour search unit 16 sequentially searches the contours of the pattern from the search start point determined by the search start point determination unit 15.

(4): The change point extraction processing section 17 extracts change points from the information obtained by the contour search. (5): The separation point candidate extraction unit 19 extracts a separation point candidate for separating characters from each other based on the change points extracted by the change point extraction processing unit 17.

(6): Based on the separation point candidates extracted by the separation point candidate extraction section 19, the size comparison section 20 obtains a circumscribed rectangle of the separated characters when the characters are separated, and the size thereof is determined. To compare.

(7): The separation line determination unit 21 determines the character and the character separation line based on the size comparison result of the size comparison unit 20. §3: Description of character cutout method ... See FIGS. 2 and 3. Hereinafter, the character cutout method will be described with reference to FIGS. In the character cutting process, when a continuous line is extracted, the connected pattern extraction unit 2, the horizontally long pattern extraction unit 3,
Continuation line extraction unit 4, character separation line determination unit 5, zero determination unit 6,
The character separating unit 7 performs character cutting processing in the same manner as in the conventional example.

When the continuation line is not extracted, after the continuation line extraction processing in the continuation line extraction unit 4 is completed, the change point extraction unit 10, the separation point determination unit 11, and the character separation unit 12 perform character cutting processing. .

In the character slicing device having the above-mentioned structure, a process is performed for extracting and separating consecutive characters from a freely written character string. Regarding characters, there is no regulation on the size of the characters or the number of characters written consecutively. Regarding the input pattern, if the slope is extreme, the slope is corrected to remove noise,
A binary image after pre-processing such as filling in a faint portion is used.

First, in order to stably pick up each pattern without depending on the relative relationship of the positions where the characters are arranged, the connected pattern extraction unit 2 uses, for example, 8
The connected patterns are extracted by labeling. Hereinafter, these are discriminated and the continuous characters and contact characters are separated. Also, the coordinates of the corners of the rectangle obtained by approximating the partial pattern obtained by the labeling to the rectangle are calculated during the labeling process.

The horizontal pattern extraction unit 3 calculates the aspect ratio of the circumscribed rectangle for each of the partial patterns obtained by the labeling as candidates for the continuous character and the contact character, and extracts the horizontal pattern having a certain value or more. To do.

The continuation line extraction unit 4 extracts a straight line by laterally adjacent projection (the projection value of a certain row is added to the projection values of surrounding rows to obtain the projection value of that row), and the line is extracted. A line with a certain length or more is extracted as a continuous line.

When the continuation line is extracted by the above processing (when there is a horizontal continuation line or an oblique continuation line), the character separation line determination unit 5, the zero determination unit 6, and the character separation unit 7 detect the zero of the number. When continuous character separation processing is performed and continuous lines are not extracted (when there is no horizontal continuous line or diagonal continuous line)
Performs the contact character separation processing by the change point extraction unit 10, the separation point determination unit 11, and the character separation unit 12.

When the continuation line is not extracted in the processing of the continuation line extraction unit 4, the change point extraction unit 10 extracts the change point from the pattern to be processed, and the separation point determination unit 11 After determining the character separation points based on the change points, the character separation unit 12 separates the characters.

In this case, first, the search start reference line determination unit 1
Reference numeral 4 determines a reference line for starting a search for searching for a change point of a continuous character, a touched character or the like when a continuous line is not extracted. Next, the search start point determination unit 15 determines the search start point based on the search start reference line determined by the search start reference line determination unit 14.

When the search start point is determined, the contour search unit 16
Search the contour of the pattern in order from the search start point.
Then, the change point extraction processing unit 17 extracts a change point from the information obtained by the contour search.

Thereafter, the separation point candidate extraction section 19 extracts the separation point candidates for separating the characters from each other based on the change points extracted by the change point extraction processing section 17, and the size comparison section 20 Based on the separation point candidates extracted by the separation point candidate extraction unit 19, the circumscribed rectangles of the separated characters when the characters are separated (when the temporary separation is performed) are calculated and their sizes are compared.

When the above processing is completed, the separation line determining section 21
Determines the character and character separation line according to the size comparison result of the size comparison unit 20. When the character separation line is determined in this manner, the character separation unit 12 separates the characters. Hereinafter, the processing of each unit will be described based on a specific example.

§4: Process description when no continuation line is extracted--see FIGS. 4 to 6 FIG. 4 is a process description diagram 1 of the first embodiment. FIG. 4A is a contour search start point, and FIG. Is an explanatory diagram of the contour search, and FIG. C is a diagram showing the inclination change point detection processing. FIG. 5 is a process explanatory diagram 2 of the first embodiment. FIG. 5A is a change point of inclination of a character pattern, FIG. 5B is a character circumscribing rectangle after character separation (part 1), and FIG. Part 2). FIG. 6 is a process explanatory diagram 3 of the first embodiment, FIG. 6A is a separation point determination, B is a character separation, and C is a character separation explanatory diagram.

Hereinafter, the processing when the continuation line is not extracted will be described with reference to FIGS. Note that this processing is performed by the change point extraction unit 10, the separation point determination unit 11, and the character separation unit 1.
2 is each processing. Further, it is assumed that the pattern to be processed is, for example, consecutive numbers 2 and 3.

(1): Description of Process of Change Point Extraction Unit First, when the continuation line is not extracted by the process of the continuation line extraction unit 4, the change point extraction unit 10 performs change point extraction processing. In this case, first, the search start reference line determination unit 14 calculates the temporary width of one character from the vertical length of the pattern of interest or the average size of other patterns, and along with that, the search start reference line L1 is roughly calculated. Is determined (see FIG. 4A).

Then, the search start reference line determining section 14 calculates the line density in the vertical direction in the peripheral portion based on the search start reference line L1 and defines the portion having the line density 1 as the formal search start reference line L2. To do.

After that, the search start point determining unit 15 determines the upper end and the lower end of the search start reference line L2 as the search start points (point 1 and point 2) (see FIG. 4A). In this case, since the contour search start point searches both upper and lower sides of the pattern, the upper and lower search start points are also set (point 1 and point 2).

Next, the contour search unit 16 performs contour search in the left and right directions of the pattern from the search start points (point 1 and point 2) obtained in the above processing (see the dotted arrow in FIG. 4B). ), The XY coordinates of the contour are stored in the internal memory.

Then, the change point extraction processing section 17 detects a change point of the inclination of the contour based on the XY coordinates of the contour obtained by the contour search and extracts the change point (see FIG. 4). (See Figure C). This process is performed as follows.

That is, by using the coordinate values of the contours already detected, the search start point (for example, point 1) and the points P1, P2, P3, P4, P on the contours distant from the search start point.
5, P6, P7, ... are sequentially connected by a straight line.

For example, a straight line connecting points 1 and P1, points 1 and P
A straight line connecting points 2, a straight line connecting points 1 and P3, a straight line connecting points 1 and P4, a straight line connecting points 1 and P5, a straight line connecting points 1 and P6, a straight line connecting points 1 and P7 ... Connect a straight line between the two points.

In this case, if the contour of the pattern from point 1 to a certain point P5 is a straight line, a straight line connecting points 1 and P1, a straight line connecting points 1 and P2, a straight line connecting points 1 and P3,
The straight line connecting the points 1 and P4 and the straight line connecting the points 1 and P5 all overlap the contour of the pattern.

However, if there is a point whose slope changes in the middle, such as after the point P6, the straight line connecting the search start point and each point after the point P6 does not overlap the contour of the pattern.
That is, if there is a pattern inclination after the point P5, the point 1
And the straight line connecting the points P5 and subsequent points deviates from the contour.
Therefore, the straight line connecting the points 1 and P6, the straight line connecting the points 1 and P7, ... Do not overlap with the contour of the pattern.

By performing the above processing, when the ratio of the overlap between the contour between two points and the straight line connecting the two points decreases, it is assumed that the point having the change in inclination has passed and the immediately preceding point is changed. Determine as a point. In this example, the point P5 is the change point.

(2): Description of Processing of Separation Point Determining Section The separation point determining section 11 performs the separation point determining processing based on the change points extracted by the change point extracting section 10 as follows.
First, the separation point candidate extraction unit 19 virtually separates the change points detected in the above process to extract the separation point candidates.

Thereafter, the size comparison unit 20 calculates the size of the circumscribed rectangles of the separated characters by the labeling process, and compares the sizes of the circumscribed rectangles so that the respective sizes are as uniform as possible. Next, the separation line determination unit 21
Determines the separating line so that the circumscribed rectangles have the same size. Specifically, the processing is performed as follows.

For example, as shown in FIG. 5A, it is assumed that points ,,,, are extracted as change points. In this case, when the circumscribing rectangle after each separation is obtained by using as the change point, the circumscribing rectangle of FIG. 5B is obtained. Further, when the circumscribed rectangle after each separation is obtained by adopting as the change point, the circumscribed rectangle of FIG. 5C is obtained.

Therefore, when the sizes of the circumscribed rectangles are compared, the size of the circumscribed rectangle of the character after separation is more uniform in the case of separating by and than in the case of separating by points. Therefore, the separation point is determined to be a point or a point.
When the upper and lower change points are close to each other like the above-mentioned change point, the separation line is determined so that the characters are separated by a straight line connecting the upper and lower change points.

However, in FIG. 6A, the change point P
5 and P6, when the distance between the search start point (point 1, point 2) and the change point is different by a certain threshold value or more (point 1
-The distance between P6 and the distance between point 2-P5 are different)
A point P5, which is a change point closer to the search start point (points 1 and 2 indicated by ●), is selected, and a virtual change point Q5 is set so as to vertically separate the pattern at the point P5. create. Then, the line connecting the points P5 and Q5 is determined as the separation line.

It is the thickness of the line segment extracted by the adjacent projection performed at the beginning of the processing that determines the concrete threshold value of the distance of the change point. Calculate the average thickness of the extracted line segments and regard that value as the average thickness of the character strokes. If the distance between the upper and lower change points is longer than the average thickness, the change point is virtually If the average thickness is less than or equal to the average thickness, the character separation line is determined so as to be separated between the upper and lower change points.

(3): Description of processing of character separation unit ... See FIG. 6 In the character separation unit 12, the character separation line determined by the above processing is used for separation. In this process, as shown in FIG. 6B,
In the case of separating between two points, when erasing with one dot line, there will be a part that is connected by eight connections.

Therefore, as shown in FIG. 6C, both sides of the erased pixel are erased at the same time (black marks are pixel erased) to prevent connection by eight connections. §5: Explanation of the processing of the change point extraction unit by a flow chart
.. Refer to FIG. 7. FIG. 7 is a processing flowchart of the change point extraction unit. Hereinafter, the processing of the change point extraction unit will be described with reference to FIG. Note that S1 to S8 indicate processing steps.

First, the search start reference line determination section 14 roughly estimates the number of characters from the aspect ratio of the pattern and determines the search start reference line L1 (S1). Then, the linear density is calculated in the vertical direction in the peripheral portion based on the search start reference line L1, and the portion having the linear density 1 is officially searched for as the reference line L2.
(S2).

Next, the search start point determination unit 15 determines the upper end and the lower end of the search start reference line L2 as the search start points (point 1 and point 2) (S3). Then, the contour search unit 16
Performs contour search in the horizontal direction of the pattern from the search start points (point 1 and point 2) obtained in the above process,
The Y coordinate is stored in the internal memory (S4).

Then, the changing point extraction processing section 17 detects a changing point of the inclination of the contour based on the XY coordinates of the contour obtained by the contour searching, and extracts the changing point. In this process, the contour line between the search start point and the point of interest and 2
The overlap with the straight line drawn between the points is calculated (S5).

As a result, if the calculated overlap is not less than the fixed threshold value (when the straight line between the two points and the contour are substantially overlapped), the process of S5 is performed, but the calculated overlap is the fixed threshold value. In the case of the following (when the straight line between the two points and the contour no longer overlap), the change point is detected (S7).

When the change point is extracted, the change point closest to the search start point is set as the character separation candidate point (S8). §6: Explanation of processing of separation point determination unit and character separation unit by flow chart ... See FIG. 8. FIG. 8 is a processing flowchart of the separation point determination unit and character separation unit. The processing of the determining unit and the character separating unit will be described. Note that S11 to S20 indicate processing steps.

Separation point candidate extraction section 19 includes change point extraction section 1
Virtual separation is performed at the change points extracted at 0, and separation point candidates are extracted. In this case, first, the separation point candidate extraction unit 19
Is another one on the same side (left or right) as the separation candidate point
The distance from one change point is calculated (S11).

Then, it is determined whether or not the threshold value calculated based on the average thickness of the character line segment is larger than the calculated distance (S12). As a result, if the threshold calculated based on the average thickness of the character line segment> the calculated distance, the separation point candidate is extracted so as to be vertically separated from the separation candidate point (S19).
However, in other cases, the separation point candidates are extracted so that the characters are separated by the line connecting the two candidate points (S13).

After the processing of S13 and S19 is completed, the size comparison unit 20 calculates the circumscribed rectangle of the separated character when the separation point candidate is virtually separated (S1).
4) The sizes of the left and right characters are compared (S15).

Next, the separation line determination unit 21 determines the separation line so that the sizes of the circumscribed rectangles are equal according to the result of the character size comparison. In this case, if the sizes of the left and right characters are different, it is not determined as a separation line (S2
0), but when the sizes of the left and right characters are substantially equal, the separation candidate point is determined as the separation point (S16). And
A character separation line is determined so as to separate the upper and lower change points determined as the separation points.

When the character separating line (straight line) connecting the two separating points is determined in the above processing, the character separating unit 12 separates the characters by the separating line (S17). Then, the pixels on both sides of the separation line (straight line) are also erased (S18).

§7: Description of Processing Results--See FIG. 9 FIG. 9 is an explanatory diagram of processing results according to the first embodiment. The result of processing by the character cutting device is as shown in the figure.

In FIG. 9, Examples 1 to 13 are examples of the character patterns separated and cut out by the character cutting device of the first embodiment. For example, Example 1 is an example in which the consecutive numbers 2 and 4 are separated, and 2 and 4 are well separated. In addition, Example 2 is an example in which the continuation of the letter B and the letter E is separated, which is well separated.

(Explanation of Second Embodiment) §1: Description of Character Slicing Device Configuration--See FIG. 10 FIG. 10 is a configuration diagram of a character slicing device according to the second embodiment. As shown in the figure, the character cutting device includes a connection pattern extraction unit 2, a continuous character extraction unit 23, a continuous line extraction unit 4, a character separation line determination unit 5, a zero determination unit 6, a character separation unit 7, and a change point extraction unit. 10, a separation point determination unit 11 and a character separation unit 12 are provided.

The connection pattern extraction unit 2, the zero determination unit 6, and the character separation unit 12 in the above configuration are the same as those in the first embodiment. The function of each part is as follows. (1): The connection pattern extraction unit 2 extracts a connection pattern from the input pattern. That is, when performing character segmentation, first, in order to stably pick up each character pattern without depending on the relative relationship of the positions where the characters are arranged, the connected pattern extraction unit 2
For example, a pattern connected by eight connections is extracted by the labeling process.

(2): The continuous character extracting section 23 performs a horizontally long pattern extracting process and a correction line extracting process. In the horizontal pattern extraction process, as in the first embodiment, the vertical / horizontal ratio of the circumscribed rectangle is calculated for each concatenated pattern obtained in the labeling process as a candidate for the continuous character, and the horizontal / horizontal ratio of a certain value or more Extract patterns with.

In the correction line extraction process, when two or more continuous lines are extracted, it is determined that the correction line is a correction line written on a character, and the correction line extraction process is performed. That is, the continuous character extracting unit 23 performs the projection process in the horizontal direction on the continuous character candidates. If there is a portion where the projection value exceeds a certain threshold value determined by the width of the pattern, it is extracted as a continuous line between the patterns. At this time, if two or more continuous lines are extracted, it is assumed that the correction line written on the character has been extracted and is excluded from the subsequent continuous character separation processing.

(3): The continuation line extraction unit 4 finds whether or not the extracted connection pattern is a continuation character by finding a continuation line of a portion where characters are connected horizontally or diagonally. Judgment is made and a horizontal or diagonal continuation line is extracted.

In this case, pattern projection processing is performed in a plurality of angular directions. Then, if there is a projection value that exceeds a predetermined threshold value, it is extracted again as a continuous line.

(4): The character separation line determination unit 5 obtains the number of characters and the area where each character exists by using contour search based on the extracted continuation line, and determines the space between characters. A character separation line such as a vertical separation line or a diagonal separation line is drawn on the.

In this case, based on the position information of the continuation line obtained by the continuation line extraction unit 4, a blank portion between characters is searched for with a broken line. When the polygonal line reaches the bottom of the pattern, it is determined that the pattern can be separated.

(5): The zero determining section 6 determines, for each character, for each character area separated by the character separation line.
It is to determine whether the number is zero (0). (6): The character separation unit 7 erases unnecessary continuation lines appearing on the left and right for the character determined to be the number zero (0), and the characters other than zero are separated by the character It is separated by a separation line. In addition, only the image above the continuation line extracted from the continuous character is used to remove the unnecessary "whisker" portion in the character that occurs when the continuous character is separated.

(7): The change point extraction unit 10 is the processing target when the continuation line (horizontal continuation line or diagonal continuation line) cannot be extracted by the processing of the continuation line extraction unit 4 (when there is no continuation line). The change point is detected for the pattern.

(8): The separation point determination unit 11 determines the character separation point based on the change points extracted by the change point extraction unit 10. (9): The character separation unit 12 performs character separation based on the character separation points determined by the separation point determination unit 11.

§2: Description of correction line determination process ... See FIG. 11. FIG. 11 is an explanatory diagram of the correction line determination process. FIG. 11A is an example of a normal continuation character, and FIGS. Is. The correction line extraction process performed by the continuous character extracting unit 23 will be described below with reference to FIG.

As described above, the continuous character extracting section 23 performs horizontal projection processing on continuous character candidates, and there is a portion where the projected value exceeds a certain threshold value determined by the width of the pattern. For example, it is extracted as a continuous line between patterns.

As described above, when extracting a continuation line, a correction line (see FIGS. B and C) may be mistakenly extracted outside the continuation line (see FIG. A) in the continuation character. In most cases, two or more correction lines are written. Therefore, if two or more continuation lines are extracted in the continuation line extraction processing, it is determined that they are correction lines. Then, the character from which the correction line is extracted is regarded as a corrected character and is excluded from the continuous character processing.

Specifically, when one continuous line is extracted as shown in FIG. A, it is determined that it is a continuous line of continuous characters, but B
When the two continuation lines are extracted as shown in FIGS. 7C and 7C, it is determined that they are not continuation lines but correction lines. Then, it is determined that the characters in FIGS. B and C are corrected characters, and they are excluded from the subsequent processing.

§3: Description of Continuation Line Extraction Processing ... Refer to FIGS. 12 and 13. FIG. 12 is an explanatory diagram of continuation line extraction processing according to the second embodiment. FIG. FIG. B is an example in which a continuous line is not extracted, and FIG. C is a diagram showing changes in the projection direction. Further, FIG. 13 is an explanatory diagram of a method for calculating a diagonal projection value.

As described above, the continuation line extraction unit 4 finds whether or not the extracted connection pattern is a continuation character by finding a continuation line of a part where characters are connected horizontally or diagonally. Judgment is made and a horizontal or diagonal continuation line is extracted.

In this case, as shown in FIG. 12C,
The pattern is obliquely projected in a plurality of angular directions. Then, if there is a projection value that exceeds a predetermined threshold value, it is extracted again as a continuous line.

The continuation line extraction unit 4 determines whether or not a continuation character candidate character is a continuation line connecting characters in order to determine whether or not it is a continuation character. Evaluate whether there is a linear component. Specifically, black pixels forming a character are projected in a certain angle direction.

If there is a continuous line, the projection value becomes large. Therefore, the presence or absence of a continuous line is evaluated depending on whether the projection value exceeds a predetermined threshold value. At this time, projection processing is performed not only in one direction but also in a plurality of directions, so that processing that does not depend on the inclination of the continuation line becomes possible.

FIG. 12A is an example of the case where the projection direction and the slope of the continuation line match and the continuation line is extracted. Further, FIG. 12B is an example in which the continuation line is not extracted because the projection direction and the inclination of the continuation line are different. Thus, by evaluating the projections in a plurality of directions, it is possible to accurately extract the continuation line.

As described above, the continuation line extraction unit 4 takes an oblique projection of a pattern for a plurality of angles, and if there is a projection value exceeding a certain threshold value, it is extracted as a continuation line. . The diagonal adjacent projection method used in this case is shown in FIG.
3 shows.

In this diagonal direction adjacent projection method, projection is performed in a certain fixed diagonal direction, and the projection value of the diagonal line is obtained by adding the projection values of the diagonal line of interest to the projection values of its surroundings. The value. The angle of the diagonal line is (the amount of change in the Y direction)
The value of / (change amount in the X direction) is set to a value that is a fraction of an integer.

In FIG. 13, an oblique projection when an n × n region is obliquely projected at an angle such that the value of (change amount in Y direction) / (change amount in X direction) is 1/2 The calculation method of the value is shown.

In this example, the projection value in the vertical direction is p1 (i).
(I = 1 to n), the projection value in the horizontal direction is p2 (j) (j = 1
˜n) and p1 (n) = p2 (1). In this case, the formulas for calculating the projection values p1 (i) and p2 (j) are as follows.

[0183]

[Equation 2]

In the above calculation formula, I (x, y)
Is a pixel value. The calculation of the diagonally adjacent projection values is performed by adding the projection values around the projection value of interest to the adjacent projection value, as in the case of the horizontal projection value.

§4: Process Description of Character Separation Line Determining Unit ... See FIG. 14 FIG. 14 is a process explanatory diagram of the character separation line determining unit, FIG. The separation determination result according to Example 1 and FIG. 6C show the separation determination result according to the method of Example 2.

The character separation line determination unit 5 obtains the number of characters and the area where each character exists by using contour search based on the continuous line extracted by the processing of the continuous line extraction unit 4, A character separation line such as a vertical separation line or a diagonal separation line is drawn between the two. In this case, based on the position information of the continuation line obtained by the continuation line extraction unit 4, a blank part between characters is searched for with a polygonal line.

In this process, for example, as shown in FIG. 14A, a character and a blank area of the character are vertically searched vertically until a black pixel is found, starting from the lower part of the continuous line area (line). Number 1). Next, the left and right black pixels are found in the horizontal direction with respect to the Y coordinate of the midpoint of the start point and the black pixel (line number 2).

Further, starting from the middle point of the left and right black pixels, a vertical search is performed until a black pixel is found in the vertical direction (line number 3). If the search position reaches the bottom of the pattern by repeating the process in the same manner, it is determined that the pattern can be separated.

Since the conventional method uses the straight line for the separation determination, it is impossible to separate the separation (see FIG. B), but by using the method of the second embodiment, the separation can be correctly determined by the broken line. (See Figure C).

§5: Process Description of Change Point Extraction Unit--See FIG. 15 FIG. 15 is a process explanatory diagram of the change point extraction unit. FIG. A is the change point process of the first embodiment, and FIG. The change point processing of is shown. For comparison, the processes of Example 1 and Example 2 are shown. Further, in the following description, the “change point” is a point where the curvature changes on the contour line, and the “contact point” is a point where characters are in contact with each other.

When the continuation line is not extracted by the process of the continuation line extraction unit 4 and when it is not determined that the number is zero by the process of the zero determination unit 6, the change point extraction unit 10 performs the extraction process of the change point. I do.

In this change point extraction processing, as shown in FIG. 15B, a point having a linear density of 1 in the vertical direction is found on the contour line within the specified range for the pattern to be processed. Next, the specified range is divided into n equal parts, and each point divided into n equal parts is used as a search start point to search for a change point in the left and right directions. To extract.

As shown in FIG. 15A, in the first embodiment, only one change point is extracted at the left and one at the maximum (see FIG. 5). Therefore, in some cases, the extracted change points do not include a point at which characters are in contact with each other (contact point).

However, in the second embodiment, as shown in FIG. 15B, the change points are extracted based on a plurality of search start points, so that more change points are extracted and contact points are extracted. Increase the probability of

At the time of recognition, a recognition process is performed on a pattern in which consecutive characters are separated at each change point and the pattern is separated at the change point having a high evaluation value, so that the contact point is included in the change points. Need to be. The processing other than the above is the same as that of the first embodiment.

§6: Description of Limitation of Search Range (Search Start Point, Change Point)-See FIG. 16 FIG. 16 is an explanatory diagram of limitation of the search range. As described above, in the change point extraction unit 10, for the pattern to be processed,
A point having a line density of 1 in the vertical direction is found on the contour line within the specified range. The designated range in this case is determined by the process of limiting the search range (search start point, change point) described below.

Generally, the contact point is likely to exist near the center of the continuous character, and its existence probability decreases toward the end. Therefore, by limiting the search start point and the search range of the change point in advance, the search process can be speeded up, and at the same time, the extraction of the change point having a low contact point likelihood closer to the edge can be reduced. It is possible to prevent an increase in the time required for the recognition processing.

In the process of limiting the search range, the height of consecutive characters,
Alternatively, the search range of the search start point and the change point is limited based on the average character size of the character string to which the continuous character belongs. For example, in the first embodiment, since the entire range of the character is set as the search range, the change point located at the end of the character is also extracted. However, in the processing of the second embodiment, unnecessary change points can be prevented from being extracted by limiting the search range as described above.

Specifically, it is as follows. : As shown in FIG. 16A, the first search range limiting process is a process of limiting the search start point and the search range based on the height of the continuous character. For example, as shown in the figure, when the height of the continuous character is ht, the search range is W, and the threshold value is A, the search range is obtained as W = ht × A.

The second search range limiting process is, as shown in FIGS. 16B and 16C, the search range of the search start point and the change point based on the average character size of the character string to which the continuous character belongs. Is a process for limiting

For example, when there is a character string as shown in FIG. 16B, the average size of the character string is calculated. Then, as shown in FIG. 16C, the character height ht is obtained from the average size. At this time, when the height of the continuous character of the average size is ht, the search range is W, and the threshold value is A, the search range is obtained as W = ht × A.

By the way, in the first embodiment, as shown in FIG. 16D, since the entire range of the character is set as the search range W, the change point (for example, the lower end of the numeral 2) located at the end of the character. ) Was also extracted.

However, since the search range W is limited in the processing of the second embodiment, extraction of unnecessary change points can be prevented as shown in FIG. 16E (for example, numeral 2).
The lower end of is not extracted as a change point).

§7: Description of Beard Removal Processing, Etc. See FIG. 17 FIG. 17 is an explanatory diagram of the beard removal processing. Character separator 7,1
In 2, the whiskers and the like of the characters generated when separating the continuous characters are removed. In this process, only the upper image of the continuous characters is stored in the memory for processing.

That is, in order to remove the beard part of the characters contained in the continuation characters and the continuation line connecting the characters,
The erase process is performed based on the image from the upper end of the pattern to the lower part of the continuation line. Specifically, of the character line segments included in the upper contour line of the pattern and the lower part of the continuation line, the lower one is removed.

For example, it is assumed that there is a continuous character of zeros as shown in FIG. In this case, B in FIG.
As shown in the figure, only the upper part of the image is stored in the memory, and processing such as beard removal is performed. By doing so, the memory capacity can be reduced as compared with the case where the entire pattern is stored in the memory and processed.

(Other Embodiments) The embodiments have been described above, but the present invention can also be implemented as follows. (1): Not limited to the handwritten character recognition device, it can be applied to the character cutting process of various recognition devices such as a print character recognition device and a drawing recognition device.

(2): The character to be cut out is not limited to the number zero, but all numbers other than zero, letters, and other various characters can be applied.

[0209]

As described above, the present invention has the following effects. (1): When a horizontal continuation line or a diagonal continuation line is not extracted, it is impossible to cut out a character with the conventional character cutting device, but in the present invention, the inclination of the character line segment is processed by the change point extraction unit. Since the change point of is extracted and the character is separated, it is possible to reliably cut out the character even when the continuation line is not extracted.

Therefore, since the contact point between characters can be accurately found, the adverse effect on character recognition can be reduced (effects corresponding to claims 1 to 4). (2): It is possible to reliably cut out characters with the same algorithm from continuous characters in which characters are continuously written to continuous characters in which characters are in contact with each other. Further, in the conventional processing, a character corrected by a double line may be erroneously processed as a continuous character. However, in the present invention, since the continuation line and the correction line can be distinguished, more accurate continuation character separation processing can be performed (effects corresponding to claims 2 and 4).

(3): By performing the contour search from the portion where the line density in the vertical direction is 1, the connection between characters can be surely found (the effect corresponding to claim 5). (4): Since the starting points of the pattern contour search are the upper and lower two points of the pattern, a total of four inclination change points can be found, that is, the upper left and right two points and the lower left and right two points. ,
Even if the contact point is shifted to the left or right, the candidate point including the correct separation point can be accurately extracted (effect corresponding to claim 6).

(5): By performing contour search to the left and right from the search start point and extracting the change points of the inclination, not only the characters are vertically separated, but even if the separation line is diagonal, It can be separated (effect corresponding to claim 7).

(6): Depending on the degree of overlap between the detected contour, or the contour and the neighboring pixels, the straight line connecting the contour search start point and a certain point on the contour. By the process of determining whether the contour up to the point is a straight line, it is possible to accurately extract the change point of the inclination of the contour even if the pattern has some irregularities ... (Effect of claim 8).

(7): The straight line connecting the starting point of contour search and a certain point on the contour and the detected contour or the overlap between the contour and its neighboring pixels are calculated, and the position of the certain point is calculated. When you move it, you can accurately extract the point where the slope of the contour that drew the arc changes sharply due to the change in the amount of change.
-(Effect of claim 9).

(8): If the distance between the upper and lower change points is less than a certain threshold value, two points are selected as the change points, and if the distance is greater than the threshold value, the search start point is selected. By selecting a close point, even a T-shaped change point portion can be accurately extracted ... (Effect corresponding to claim 10).

(9): The thickness of the line segment extracted by the adjacent projection method in which the result of the addition of the surroundings as the reference of the distance between the upper and lower change points is used as the projection value of the target row or column. By setting it based on the average value, it is possible to deal with changes in the thickness of the lines that make up the characters.
(Effect of claim 11).

(10): Since the height of a character is almost constant depending on the size of the character, regardless of the number of connected characters, the reference for the distance between the change points of the upper and lower two points is calculated based on the height of the pattern of interest. With this value, the threshold value can be obtained without being influenced by the number of connected characters (effect corresponding to claim 12).

(11): By eliminating the straight line connecting the separation points and the pixels on both sides of the straight line, it is possible to prevent the pixels from being connected by 8 connections and the characters from being separated from each other. (Effect of claim 13).

(12): It is assumed that the characters have been separated by the separation point candidate, and the size of the character after separation is obtained by labeling, contour search, or projection processing. It is possible to calculate the size ... (Effect of claim 14).

(13): Separation points can be determined from a plurality of separation point candidates by deciding the separation points so that the calculated sizes after separation are equal. (Effect of Item 15).

(14): In the conventional processing, a character corrected by a double line may be erroneously processed continuously. However, in the present invention, since the continuation line and the correction line can be distinguished, more accurate continuation character extraction processing can be performed (effect corresponding to claim 16).

(15): When extracting the continuation line from the continuation character candidate character, if only the projection process in one direction is performed, and if that direction and the inclination direction of the continuation line are deviated, the continuation line is accurately Could not be extracted. However, according to the present invention, since projection processing is performed in a plurality of directions, it is possible to accurately extract a continuous line regardless of the inclination of the continuous line ...
(Effect of claim 17).

(16): When judging whether or not continuous characters can be separated,
Judging whether or not a straight line can be drawn between the characters, if the blank area between the touched characters becomes complicated, there is no choice but to make them inseparable.

However, according to the present invention, since it is determined whether or not a polygonal line is drawn between the touched characters, it is possible to accurately separate and determine continuous characters that were erroneously determined to be impossible by the conventional processing. It is possible ... (effect corresponding to claim 18).

(17): In the change point extraction processing, the consecutive characters cannot be accurately separated unless the points where the characters touch each other are extracted. However, according to the present invention, the search for the change points is performed. By performing the process from a plurality of starting points, the change points can be extracted in a wider range (claim 19).
Corresponding effect).

(18): The point where characters touch each other is likely to be located near the center of successive characters, and the probability of existence decreases toward the ends. If the starting point is set with the entire continuous character set as the search range, the number of starting points will increase. According to the present invention, a region having a low contact point existence probability is excluded from the starting point range when changing points are searched, so that the number of set starting points can be reduced. As a result, the processing speed can be increased.
.. (Effect of claim 20).

(19): When the change points are extracted, the change points are extracted in addition to the contact points between the characters. When the recognition processing is performed for each change point and the cutout is evaluated, the recognition processing takes longer as the number of change points increases.

According to the present invention, since the region having a low contact point existence probability is excluded from the search range, the number of change points extracted as candidates is reduced, and the processing speed can be increased.
At the same time, the search time can be shortened (effect corresponding to claim 21).

(20): When a fixed value such as the height of the character frame is used in determining the search start range of the change point, the processing result is not stable when the character size changes. According to the present invention, it is possible to set the search start range according to the character size of the pattern of interest (effect corresponding to claim 22).

(21): When a fixed value such as the height of the character is used when determining the search start range of the change point, the processing result is not stable when the character size changes. According to the present invention, it is possible to set the search range according to the tendency of the character size of the entire character string, for example, the number of horizontally long characters is large (the effect corresponding to claim 23).

(22): When determining the change point search range,
If a fixed value such as character height is used, the processing result will not be stable if the character size changes. According to the present invention, the search range can be set according to the character size ...
-(Effect of claim 24).

(23): If a fixed value such as the height of the character is used as the search range of the change point, the processing result is not stable when the character size changes. According to the present invention, the search range can be set according to the tendency of the character size of the entire character string, for example, the number of horizontally long characters is large (the effect of claim 25).

(24): If the memory capacity of the hardware is limited, it may not be possible to capture all one character in the memory. In such a case, according to the present invention,
The processing can be performed only with the partial image of the character, and the processing can be performed with a small-capacity memory ... (Effect corresponding to claim 26).

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is a basic configuration diagram of a character cutting device according to the first embodiment.

FIG. 3 is a configuration diagram of a character cutting device according to the first embodiment.

FIG. 4 is a process explanatory diagram 1 of the first embodiment.

FIG. 5 is a process explanatory diagram 2 of the first embodiment.

FIG. 6 is a process explanatory diagram 3 of the first embodiment.

FIG. 7 is a processing flowchart of a change point extraction unit according to the first embodiment.

FIG. 8 is a processing flowchart of a separation point determination unit and a character separation unit according to the first embodiment.

FIG. 9 is an explanatory diagram of a processing result of the first embodiment.

FIG. 10 is a configuration diagram of a character cutting device according to a second embodiment.

FIG. 11 is an explanatory diagram of a correction line determination process according to the second embodiment.

FIG. 12 is an explanatory diagram of a continuous line extraction process according to the second embodiment.

FIG. 13 is an explanatory diagram of a diagonal projection value calculation method according to the second embodiment.

FIG. 14 is a process explanatory diagram of a character separation line determination unit according to the second embodiment.

FIG. 15 is an explanatory diagram of processing of a change point extraction unit according to the second embodiment.

FIG. 16 is an explanatory diagram of limitation of a search range according to the second embodiment.

FIG. 17 is an explanatory diagram of a beard erasing process according to the second embodiment.

FIG. 18 is a configuration diagram of a conventional character cutting device.

FIG. 19 is a process explanatory diagram of a conventional connected pattern extraction unit, a horizontally long pattern extraction unit, and a continuous line extraction unit.

FIG. 20 is an explanatory diagram of the processing of the conventional continuous line extracting unit (No. 1).
Is.

FIG. 21 is an explanatory diagram of a process of a conventional continuous line extracting unit (part 2).
Is.

FIG. 22 is an explanatory diagram of the processing of the conventional continuous line extracting unit (Part 3).
Is.

FIG. 23 is an explanatory diagram of processing of a conventional character separation line determination unit.

FIG. 24 is an explanatory diagram of processing of a conventional zero determination unit.

FIG. 25 is an explanatory diagram of processing of a conventional character separating unit.

[Explanation of symbols]

 2 Connected pattern extraction unit 3 Horizontal pattern extraction unit 4 Continuation line extraction unit 5 Character separation line determination unit 6 Zero determination unit 7 Character separation unit 10 Change point extraction unit 11 Separation point determination unit 12 Character separation unit 23 Continuation character extraction unit

Claims (26)

[Claims] 1. A character slicing device for recognizing characters, wherein a connection pattern extraction unit extracts a connection pattern from an input pattern based on connection information of the input pattern, and a connection pattern extracted by the connection pattern extraction unit. Of the pattern
As a continuation character candidate in which a plurality of characters are connected, a horizontal pattern extraction unit that extracts only a horizontally long horizontal pattern and a continuous line that connects characters from the horizontal pattern extracted by the horizontal pattern extraction unit are found. By
It is determined whether or not the extracted connection pattern is a continuation character, and a continuation line extraction unit that extracts a continuation line of a character, and if the continuation line cannot be extracted by the continuation line extraction unit,
Based on the change points extracted by the change point extraction unit for performing pattern contour search to extract the change points of the inclination of the pattern contour, and the change points as character and character separation point candidates. , The size of the character after separation when separated by each separation point candidate is determined, and the separation point determination unit that determines the separation point of the character by comparing the sizes of the characters, and the separation point determination unit. A character slicing device having a character separation section for separating characters at character separation points. 2. A character slicing device for recognizing characters, wherein a connection pattern extraction unit extracts a connection pattern from an input pattern based on connection information of the input pattern; and a connection pattern extracted by the connection pattern extraction unit. Of the pattern
Horizontal pattern extraction processing that extracts only horizontally long horizontal patterns as candidates for continuous characters in which multiple characters are connected,
And a continuation character extraction unit that performs a correction line extraction process that extracts a continuation line connecting the characters from the horizontal pattern and extracts a correction line written on the character, and the continuation character extraction unit extracts the correction line extraction process. A continuation line extracting unit that determines whether or not the extracted connection pattern is a continuation character by finding a continuation line connecting characters from a horizontal pattern, and a continuation line extraction unit that extracts a continuation line of a character, and the continuation When the continuation line is extracted by the line extraction unit, based on the position information of the continuation line, the character separation line determination unit that determines the character separation line and the character separation line determined by the character separation line determination unit are separated. For the character area of each character, for each character, a zero determination unit that determines whether it is a numeral zero, and for the character that is determined to be zero of the numeral, leave a character area and erase unnecessary continuation parts. Character separation part to When the continuation line cannot be extracted by the continuation line extraction unit, and when the zero determination unit does not determine that the number is zero, the contour of the pattern is searched to extract a change point of the inclination of the pattern contour, A change point extraction unit that uses the change point as a character and a character separation point candidate, and the size of the character after separation when separated by each of the separation point candidates based on the change points extracted by the change point extraction unit. And a separation point determination unit that determines the separation point of the character by comparing the size of the character, and separates the character at the separation point of the character determined by the separation point determination unit, leaving a character region and unnecessary continuation. A character slicing device having a character separating section for erasing a part. 3. A character segmentation method for performing character recognition processing, comprising: a connection pattern extraction process of extracting a connection pattern from an input pattern based on connection information of the input pattern; and a connection pattern extracted by the connection pattern extraction unit. Of the pattern
A horizontal pattern extraction process that extracts only a horizontally long horizontal pattern as a candidate for a continuous character in which a plurality of characters are connected, and a continuous line that connects characters from the horizontal pattern extracted by the horizontal pattern extraction unit is found. By
A judgment is made as to whether or not the extracted connection pattern is a continuation character, and a continuation line extraction process for extracting a continuation line of the character, and a contour of the pattern when the continuation line cannot be extracted by the continuation line extraction process. A change point of the inclination of the pattern contour is extracted by performing a search, a change point extraction process in which the change point is a character and a character separation point candidate, and a change point extracted in the change point extraction process is used. Separation point determination processing for determining the size of the character after separation when separating with the separation point candidate, and character separation determined by comparing the character sizes, and character separation determined by the separation point determination processing A character segmentation method for separating characters at points. 4. A character segmentation method for performing character recognition processing, comprising: a connection pattern extraction process for extracting a connection pattern from an input pattern based on connection information of the input pattern; and a connection pattern extracted by the connection pattern extraction unit. Of the pattern
A horizontal pattern extraction process that extracts only a horizontally long horizontal pattern as a continuous character candidate in which a plurality of characters are connected, and a continuous line that connects characters from the horizontal pattern is extracted to write on the character. The continuous character extraction process consisting of the correction line extraction process for extracting the corrected line, and by finding the continuous line connecting the characters from the horizontal pattern extracted by the continuous character extraction process, the extracted connection pattern is If a continuation line is extracted in the continuation line extraction process of determining whether or not the continuation character is a continuation character and extracting the continuation line of the character, and if the continuation line is extracted by the continuation line extraction process, the character segment is extracted based on the position information of the continuation line. Character separation line determination processing for determining the separation line, for the character area for each character divided by the character separation line determined in the character separation line determination processing, for each character, whether the number is zero or not A zero determination process that determines the number, a first character separation process that leaves a character area and erases unnecessary continuation parts of the character that is determined to be zero, and a continuation line cannot be extracted by the continuation line extraction process. If
And, when it is not determined that the number is zero in the zero determination process, the change point of the inclination of the pattern outline is extracted by performing the contour search of the pattern, and the change point is a separation point candidate of the character and the character. Extraction process, based on the change point extracted in the change point extraction process, to determine the size of the character after separation when separated by each separation point candidate, the character separation point by comparing the size of the character And a separation point determining process for determining the character separation, and character separation at the character separation point determined by the separation point determining process, leaving a character region and erasing unnecessary continuation parts.
A character segmentation method characterized by having the character separation process described in 1. 5. In the change point extraction processing, after a temporary search start reference line is determined from the vertical length of the pattern of interest, the average size of other patterns, etc. 4. The character segmentation method according to claim 3, wherein a formal search start reference line is obtained by searching a portion having a line density of 1, and a search start point is determined from the search start reference line to perform contour search of the pattern. . 6. The change point extraction process according to claim 3, wherein the search start points when the contour search of the pattern is performed are two points of the upper end and the lower end of the pattern in the portion having the linear density of 1. Character cutting method. 7. The change of the inclination of the pattern by determining the search start point for starting the contour search of the pattern in the change point extraction processing and then performing the contour search in the left and right directions of the pattern from the search start point, respectively. The character cutting method according to claim 3, wherein points are extracted. 8. In the change point extraction processing, when a contour search of a pattern is performed to extract a change point of the inclination of the pattern contour, a straight line connecting a search start point on the pattern and a certain point on the contour is formed. , Calculating an overlap with the detected contour, and extracting a change point by determining whether or not a point from the search start point to a certain point on the contour is a straight line based on the calculated degree of overlap. The character cutting method according to claim 3, wherein. 9. A straight line connecting a search start point on a pattern and a certain point on the contour when the contour of the pattern is searched to extract a change point of the inclination of the contour of the pattern in the change point extraction processing. The character cutting method according to claim 3, wherein the overlap with the contour is calculated, and the change point of the inclination of the contour is extracted from the change amount of the overlap. 10. In the separation point determination process, if the distance between the change points of the upper and lower two points of the pattern is less than or equal to a certain threshold value, the two points are selected as the change points,
The character cutting method according to claim 3, wherein when the distance exceeds a certain threshold value, one point closer to the search start point is selected as a change point. 11. A threshold value of a distance between the upper and lower two points,
11. The method according to claim 10, wherein the result obtained by adding up the projection values of the surroundings is used as the projection value of the target row or column, and the average value of the thicknesses of the line segments extracted by the adjacent projection method is set as a reference. Character cutting method. 12. A threshold value of the distance between the upper and lower two points,
The character cutting method according to claim 10, wherein the height is set based on the height of the pattern of interest. 13. In the character separation process, when a character is separated at a character separation point determined by the separation point determination unit, a straight line connecting the separation points and pixels on both sides thereof are erased. 4. The character cutting method according to claim 3. 14. In the separation point determination process, it is assumed that the separation points are separated by respective separation point candidates based on the change points extracted by the change point extraction unit, and after separation by labeling, contour search, projection method, or the like. 4. The separation point is determined by determining the size of the character of "."
How to cut out the described characters. 15. In the separation point determination process, it is assumed that each separation point candidate is separated based on the change point extracted by the change point extraction unit, the size of the character after separation is obtained, and the separation is performed. The character cutting method according to claim 3, wherein the separation points are determined so that the subsequent sizes are equal. 16. The continuous character extracting process, when extracting a continuous character from a continuous character candidate, performs horizontal projection processing to extract two continuous lines, and the two continuous lines are referred to as correction lines. The character segmentation method according to claim 4, wherein the character is judged and the character is treated as a corrected character. 17. In the continuation line extracting process, when extracting a continuation line connecting characters from the horizontally elongated pattern extracted in the continuation character extracting process, the continuation line extracting process is performed for a plurality of slopes. Claim 4 characterized by the above.
How to cut out the described characters. 18. In the character separation line determination process, the continuation line extraction unit extracts a continuation line, and when the character separation line is determined based on position information of the continuation line, is it possible to separate two characters by a polygonal line? The character cutout method according to claim 4, wherein whether or not the character can be separated is determined depending on whether or not the character is separated. 19. In the change point extraction process, when extracting a change point as a point of contact between characters, first, a point on the contour line having a line density of 1 is found in the vertical direction, and then a change point is searched for. The character cutout method according to claim 4, wherein a plurality of start points are set on the contour line, and the plurality of change point candidates are calculated from each point. 20. The character cutting method according to claim 19, wherein when a plurality of search start points of the change points are set on the contour line, the search start point is found by limiting the search range for the pattern. 21. When calculating the plurality of change point candidates,
20. The character cutout method according to claim 19, wherein a change point candidate is calculated by limiting a search range for the pattern. 22. The character cutting method according to claim 20, wherein when the search range of the search start point is limited, a constant multiple of the height of the pattern is set as the search range. 23. The character cutting method according to claim 20, wherein, when limiting the search range of the search start point, a constant multiple of the average character size of the character string to which the character belongs is set as the search range. 24. When limiting the search range of the change point,
22. The character cutting method according to claim 21, wherein the search range is set to a constant multiple of the height of the pattern. 25. When limiting the search range of the change point,
22. The character cutting method according to claim 21, wherein the search range is a constant multiple of the average character size of the character string to which the character belongs. 26. In the first and second character separation processes, while using only an image in a rectangle surrounding a continuation line to remove a beard of a character caused by separation of continuation characters, The character cutting method according to claim 4, wherein a beard portion included in the character is removed from position information of the contour line and the continuation line on the upper side of the pattern.