JPH03244094A - Character recognition device - Google Patents

Abstract

PURPOSE:To obtain a character recognition device which is fast and small in size and provides good operability by providing a subpattern extracting means, a character direction decision means, a converting means, and a rotation direction decision means. CONSTITUTION:The subpattern extracting means 13 extracts the vertical subpattern and horizontal subpattern of an input pattern and the character direction decision means 20 decides whether the input pattern is written longitudinally or laterally, i.e. the character direction according to the extraction result. Then the converting means 15 extracts two different features from the input pattern and the rotation direction decision means 23 decides the angle of rotation of the character direction decision result outputted by the character direction decision means 20 according to the results obtained by discriminating the extracted features by a discriminating means 22 and decides the character direction of the input pattern. Therefore, a dictionary for four-directional matching need not be prepared and the storage capacity of a dictionary storage means 21 is reducible. Consequently, the character recognition device is reduced in size and the processing time may be short; and the processing speed can be increased and the operability is improved.

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial Field of Application) The present invention relates to an optical character reading device that optically reads printed documents, etc.
The present invention relates to a high-speed, small-sized, and easy-to-operate character recognition device such as OCR (hereinafter referred to as OCR).

(Prior art) Conventionally, as a technology in this field, Japanese Patent Application Laid-Open No. 62-16
There was one described in Publication No. 6479.

In recent years, with the spread of Japanese information processing systems, the demand for inputting Japanese printed documents into computers has rapidly increased, and the need for printed kanji OCR has rapidly increased to speed up input and save labor. .

This printed kanji OCR is equipped with a character recognition device that reads kanji, etc. in a printed document, and
It is necessary to identify the character pattern of the read kanji.

By the way, printed documents include horizontally written documents and vertically written documents such as newspapers and paperback books.

For example, if we take a business card as an example of a printed document, as shown in Figure 2 (a-1>, (a-2) to (di,) (d-2>), horizontally written business cards (Figure 2 (a) -1>, (b-
King)> and vertical business cards (Fig. 2 (c-1), (d
-1>). Furthermore, depending on how the business card is placed, even if the form of the form is the same, it may be rotated by -80°.

For example, Fig. 2 (b↑) is the state in which Fig. 2 (a-1> is rotated by 1-80 degrees, and Fig. 2 (c-1,> is rotated by 1-80 degrees).
FIG. 2(d-1) shows the state rotated by .degree. Also, the second
Figures (a-1,), (b-1>, (c-1>, (d
Examples of each pattern of l) are shown in Fig. 2 (a-2>, 12) and (c
-2> and (d-2), respectively. here,
Based on Figure 2 (a-2>), Figure 2 (b-2) is 1
-80°, same figure (c-2> is 270°, same figure (d-2)
is a pattern rotated clockwise by 90 degrees.

In the conventional character recognition device for printed kanji 0CRJ, the input pattern read by the photoelectric conversion means is stored in the pattern storage means, and the feature extraction means extracts the characteristics of the input pattern from the input pattern stored in the pattern storage means. do. Then, the identification means identifies the input pattern by comparing the features extracted by the feature extraction means with a dictionary stored in the dictionary storage means.

In such a character recognition device, as shown in Figure 2, printed documents have vertical and horizontal writing, and forms have 180 characters.
°When recognizing a document that may be set in a rotated position, character recognition cannot be performed correctly unless the operator specifies the character direction for each character or form using a keyboard while looking at the form. Therefore, in order to solve such a problem, the technique of the above-mentioned document uses four
A dictionary of types is stored in a dictionary storage means in advance. Then, the input pattern is recognized with the four types of dictionaries, and from the recognition results obtained by each dictionary, the direction of the dictionary that outputs the result with the highest degree of similarity is determined to be the character direction,
Characters were recognized based on the determination results.

(Problems to be Solved by the Invention) However, the apparatus with the above configuration has the following problems.

(a> 0°, 90″180 for one character pattern
Since it is necessary to perform identification four times corresponding to rotations of 270° and 270°, there is a problem in that character recognition processing takes time.

(b) Since four types of dictionaries such as 06.90°, 180', and 270' must be prepared, the memory capacity of the dictionary storage means for storing them becomes extremely large, leading to an increase in the size of the device. There was a problem.

(c) In the technology disclosed in the above document, the character direction is determined and character recognition is performed based on the determination result, so the operator does not need to specify the character direction for each character or for each form, thereby reducing the complexity of the operation. This solves the problem of processing time and processing time. However, although the technology disclosed in the above document can automatically recognize vertically or horizontally written business cards, it cannot recognize printed documents other than business cards, and as a result, the device has low versatility. It was not technically satisfactory.

The present invention provides a character recognition device that solves the problems of the prior art, such as the recognition process taking a long time, the device becoming large, and lack of versatility.

(Means for Solving the Problems) In order to solve the above problems, the present invention provides a pattern storage means for storing read input patterns, and a feature of the input pattern stored in the pattern storage means. A character recognition device comprising: a feature extraction means for extracting features; a dictionary storage means for storing a dictionary; and an identification means for identifying the input pattern by comparing the features extracted by the feature extraction means with the dictionary. sub-pattern extraction means for scanning the input pattern stored in the pattern storage means in a plurality of directions to extract vertical sub-patterns and horizontal sub-patterns;
The apparatus is provided with a character direction determining means, a converting means, and a rotation direction determining means.

Here, the character direction determining means calculates the line widths of the vertical sub-pattern and the horizontal sub-pattern extracted by the sub-pattern extracting means, and compares the ratio of both line widths with a predetermined line width ratio reference value to determine the character direction of the character. The conversion means has a function of determining the direction (for example, horizontal writing, vertical writing).The conversion means performs feature conversion or pattern conversion based on the output of the character direction determination means, thereby storing the text in the pattern storage means. The rotation direction determining means has a function of extracting features in two different directions from the input pattern and having the identification means identify the extraction results.Furthermore, the rotation direction determining means is configured to extract features in two different directions from the input pattern and identify the extraction results with the identification means.
The output of the character direction determining means is set at a predetermined angle (for example, O'
The character pattern has a function of determining whether the character pattern is rotated (or 180°) and determining the direction of rotation of the character pattern.

(Operation) According to the present invention, since the character recognition device is configured as described above, when characters etc. on a form are read by the photoelectric conversion means etc., the read input pattern is stored in the pattern storage means. Then, the sub-pattern extraction means scans the input pattern stored in the pattern storage means in multiple directions (for example, horizontal and vertical directions), extracts vertical sub-patterns and horizontal sub-patterns, and applies the extraction results to the character direction determination means. The character direction determining means calculates the line width of each of the extracted vertical sub-patterns and horizontal sub-patterns, calculates the ratio of the line width of the vertical sub-pattern to the line width of the horizontal sub-pattern, and calculates the ratio between the line width of the vertical sub-pattern and the horizontal sub-pattern. The width ratio reference value is compared to determine the character direction (for example, horizontal writing, vertical writing) of the input pattern, and the character direction determination result is provided to the conversion means and rotation direction determination means.

The converting means converts the input pattern stored in the pattern storage means into two different directions by rotating the characteristic 71 helix or input pattern of the character pattern extracted by the feature extraction means by a predetermined angle based on the character direction determination result. Extraction of features (for example, in the case of horizontal writing, feature extraction for a pattern that is not rotated and feature extraction for a pattern rotated by 80 degrees, and for vertical writing, feature extraction for a pattern rotated by 90 degrees, and 2)
70° rotated pattern), and the extraction result is identified by the identification means. The rotation direction determining means determines whether the character direction determination result of horizontal writing or vertical writing outputted from the character direction determining means is rotated by a predetermined angle (for example, 0° or 380') based on the identification result of the identifying means. The direction of rotation of the input pattern is determined by determining whether

In this way, since the character direction determining means and the rotation direction determining means each determine the character direction, the number of times of identification for one character pattern is reduced, and the recognition process can be performed in a short time. The number of dictionary storage means can be reduced, and the capacity of the dictionary storage means can be reduced, thereby making it possible to downsize the device. Furthermore, since this device can process printed documents other than business cards, its versatility can be improved. Therefore, the above problem can be solved.

(Embodiment) FIG. 1 is a functional block diagram of a character recognition device showing an embodiment of the present invention.

Before explaining this functional block diagram, the principle of this embodiment will be explained first.

Focusing on the line width of Mincho fonts, which account for most of the type used in printed documents, we calculated the ratio of the line width of horizontal sub-patterns to the line width of vertical sub-patterns, which was obtained using the sub-pattern extraction method described later. The results of the measurements are shown in Table 1.

0 Table 11 Measurement results of line width ratio of horizontal sub-pattern and vertical sub-pattern Now, regarding the character “A” in the Mincho typeface shown in Figure 3 (a>),
For type size 8.12 and ]6 point I,
Obtained by image scanning with a resolution of 37.5 μm f
A horizontal sub-pattern and a vertical sub-pattern were extracted from the s binary pattern, and the horizontal sub-pattern line width and vertical sub-pattern line width were determined, respectively.

The method for extracting horizontal sub-patterns and vertical sub-patterns is to obtain the line width of an input pattern using the (Co) formula described later, and then scan the input pattern in the horizontal and vertical directions to find a cross-section of the character line. This is performed by extracting a cross section whose cross-sectional length is sufficiently longer than the line width of the input pattern.

An example of the vertical sub-pattern of the horizontally written "A" in the Mincho font shown in FIG. 3(a) is shown in FIG. 3(b), and an example of the horizontal sub-pattern is shown in FIG. 3(C). Also, Figure 3 (a
The vertical writing pattern examples corresponding to >, (b), and (c) are shown in the third example.
These are shown in figures (d), (e) and (f), respectively. That is, FIG. 3(d) shows the vertically written Mincho font "A", which is obtained by rotating the character pattern of FIG. 3(a) by 90 degrees counterclockwise. Therefore, FIG. 3(d) and FIG. 3(e) showing the vertical sub-pattern of Kei 1-2 and FIG. 3(f> showing the horizontal sub-pattern) are respectively shown in FIG.
) is rotated 90° counterclockwise to create the same pattern.

As is clear from the table, the horizontal sub-pattern line width and the vertical sub-pattern line width of Mincho fonts differ greatly. Therefore, the line width ratio of horizontally written characters (-vertical sub-pattern line width/horizontal sub-pattern line width) is 2.8 or more.Similarly, the line width ratio of vertically written characters (-vertical sub-pattern line width) is 2.8 or more. /horizontal subpattern line width> is 0.36 or less.

As can be understood from the above, it is possible to determine whether the input pattern is horizontal writing or vertical writing, depending on the magnitude of the line width ratio (-vertical sub-pattern line width/horizontal sub-pattern line width) of the input pattern.

Next, the configuration of the character recognition device shown in FIG. 1, which is configured based on the above principle, will be explained.

This character recognition device has a photoelectric conversion means that converts characters, figures, etc. on an input form into electrical signals by photoelectric conversion and outputs a character figure pattern. - A pattern storage means 2 composed of an access memory (RAM) or the like is connected to a line width calculation means 3. The pattern storage means 2 stores, for example, one character/figure pattern outputted from the photoelectric conversion means, and the feature extraction means 10 is connected to the output side thereof. The line width calculation means 3 is a circuit that calculates the line width W of the input pattern upon receiving the character/figure pattern output from the photoelectric conversion means 1, and the feature extraction means 1O is connected to the output side thereof.

The feature extraction means 10 extracts the features of the large pattern output from the pattern storage means 2 and outputs two types of feature matrices. It is composed of extraction means 13, feature matrix extraction means 14, and feature conversion means 15. The character frame detection means 1 has a function of detecting a character frame circumscribing a character pattern in the pattern storage means 2 and outputting the detection result to the character frame division determining means 12. The character frame division determining means 12 divides the detected character frame improvement region into MXN regions (M and N are integers, and in this embodiment, for example, M=N=5). It has a function of determining division point coordinates on the X-axis (horizontal direction of the character frame) and Y-axis (vertical direction of the character frame) for the purpose of the character frame, and outputting the determination results to the feature matrix extraction means 14.

The sub-pattern extraction means 13 scans the input pattern stored in the pattern storage means 2G in a plurality of directions (for example, vertical scanning, horizontal scanning, 45° diagonal scanning to the right, and 45° diagonal scanning to the left) to obtain a vertical sub-pattern VSP. , horizontal sub-pattern H3
P, right diagonal sub-pattern R3P, and left diagonal sub-pattern L
It has a function of extracting each SP and outputting the extraction results to the feature matrix extraction means 14. The feature matrix extraction means 14 has a function of extracting a feature matrix from the output of the sub pattern extraction means ↑3 based on the division point coordinates output from the character frame division determination means 12, and providing the extraction result to the feature conversion means 1-5. have. The feature converter 15 has a function of rotating the input feature matrix in a predetermined direction and outputting two types of feature matrices.

On the output side of the sub-pattern extracting means, a character direction determining means 2 is provided.
0 is connected, and the output side of the character direction determining means 20 is
It is connected to the feature conversion means 1-5 and the rotation direction determination means 23. Furthermore, two dictionary storage means are provided, and the dictionary storage means 21 and the feature conversion means 1-5 are connected to the identification means 22.
The identification means 22 is connected to the rotation direction determination means 23
It is connected to the.

The character direction determining means 20 has a function of determining the character direction of the input pattern from the output of the sub-pattern extracting means 13 and outputting the character direction determination result to the feature converting means 15 and the rotation direction determining means 23. The dictionary storage means 21- stores, for example, a collation dictionary mask for horizontal writing, and is used for reading and writing.
It consists of only memory (ROM) and the like. The identification means 22 refers to the contents of the dictionary storage means 2]- and determines the rotation direction based on the category name and distance (similarity) of the dictionary mask for the two types of one-temple signs 71 to rikus output from the feature conversion means 15. It has a function of outputting to the means 23. The rotation direction determining means 23 has a storage means such as a buffer for storing the output of the identifying means 22, and based on the output rotation of the identifying means 22, the character direction determination result output from the character direction determining means 20 is rotated at a predetermined angle (e.g. , ○° or engineering 8
It has a function of determining whether the input pattern is rotated by 0°>, determining the direction of rotation of the input pattern, and outputting the character name and character direction, respectively.

Next, we will explain the operation of the character recognition device configured as above.
This will be explained with reference to FIGS. 4(a) to 4(e). Note that FIGS. 4(a) to 4(e) are diagrams for explaining the binary pattern and each sub-pattern.

In the above figure, when the optical signal corresponding to the characters and figures on the form is input to the photoelectric conversion means ■, the photoelectric conversion means
The input optical signal is converted into an electrical signal and a character/figure pattern is output to the pattern storage means 2 and the line width calculation means 3. The pattern storage means 2 stores, for example, the ↑ pattern of the character/figure patterns output from the photoelectric conversion means 1.

When the line width calculation means 3 receives the character/figure pattern outputted from the photoelectric conversion means ↓, it calculates the line width W of the input pattern, and outputs the calculation result to the subpattern extraction means 13 and the feature matrix extraction means 14. .

The line width calculation means 3 has a shift register configuration, for example, similar to a known image processing filter circuit, and calculates the line width W using a known approximate expression as shown in the following equation.

W-1/(1-(Q/A)+ ・・・・・・(1
) In this formula <1>, Q is a (2x2) window (Wln
dOw is the number of points in which all points in the rectangular area are black points, which is called 4 black points, and A is the total number of black points. The line width calculating means 3 counts the number Q of these four black dots and the number A of all black dots, and calculates the line width W from the counting results according to formula (1).

Next, in the feature extraction means 10, the character frame detection means 11
As a result, a character frame circumscribing the character pattern in the pattern storage means 2 is detected. The character frame division determining means 1-2 divides the internal area of the detected character frame into MXN (
For example, the X axis, Y axis for dividing into regions of M=N=5>
The coordinates of the division points on the axis are determined, and the determination results are output to the feature matrix extraction means 14.

On the other hand, in the sub-pattern extraction means 13, the pattern storage means 2
Based on the relationship between the length of continuous black dots (character line portions are black dots) by performing vertical scanning over the input pattern stored in the input pattern and the line width W calculated by the line width calculation means 3,
Extract the vertical sub-pattern VSP. Similarly, the sub-pattern extracting means 13 extracts a horizontal sub-pattern H by horizontal scanning.
A right diagonal sub-pattern R3P is extracted by scanning S P diagonally at 45 degrees to the right, and a left diagonal sub-pattern LSP is extracted by scanning S P diagonally at 45 degrees to the left.

An example of sub-patterns extracted in this way is shown in Figure 4 (a
) to (e). Figure 4(a) is an original pattern consisting of a binary pattern within the circumscribed character frame F, and the vertical double pattern obtained from such an original pattern is shown in Figure 4(a).
b), the horizontal sub-pattern is shown in Fig. 4(C), the right diagonal sub-pattern is shown in Fig. 4(d), and the left diagonal sub-pattern is shown in Fig. 4(d).
e), respectively. Such sub-patterns are provided to the feature matrix extraction means 14 and the character direction determination means 20.

The feature matrix extraction means 14 includes the character frame division determination means 1
Based on the dividing point coordinates determined in step 2, vertical, horizontal, right diagonal, and left diagonal sub-patterns (VSP, H8P,
The character frame area on the pattern storage means 2 in R3P and LSP) is divided into MXN areas, and the number of black dots Bjj in each of the resulting divided areas is counted. Using this number of black dots Bij and the line width W obtained by the line width calculation means 3, a feature indicating the character line length Lij is calculated by the following formula (2), and M×NX4
Create a dimensional feature matrix.

Lij=Bij/W (2) Furthermore, this feature matrix extraction means 14 extracts 1 0. For the feature matrix of the vertical sub-pattern VSP, the length ΔY of the character frame F in the Y-axis direction is The feature matrix of the horizontal sub-pattern H3P is normalized by the length △X in the X-axis direction, and the feature matrix of the right diagonal sub-pattern R3P and left diagonal sub-pattern LSP is normalized by (△X + △Y)/2. MX that absorbs fluctuations due to font size
Create a NX4-dimensional feature matrix.

Similarly to the line width calculation means 3, the character direction determination means 20 scans the vertical sub-pattern stored in the sub-pattern extraction means]-3, calculates the number of four black dots Qv and the total number of black dots Av of the vertical sub-pattern, -> Find the line width Wv of the vertical sub-pattern using the formula.Furthermore, scan the horizontal sub-pattern stored in the sub-pattern extraction means 13, find the number Qh of four black dots and the number Ah of all black dots of the horizontal sub-pattern, and calculate (1 ) formula to find the line width Wh of the horizontal sub-pattern.Then, calculate the line width ratio Wv/wh of the vertical sub-pattern and the horizontal sub-pattern, and calculate the line width ratio Wv/Wh and the preset line width ratio reference value. wth and determines whether the following expressions (3> and (4) are satisfied or not: W v /W h ≧wth
・・・・・・(3) Wv/Wh <Wt
h ・=−(4) (in this example, for example, Wt
h= 1 > In other words, the character direction determining means 20 determines (3
) If the formula is satisfied, the character direction is determined to be horizontal writing, and the character direction determination result for horizontal writing is determined.If the formula (4) is satisfied, the character direction is determined to be vertical writing, and the character direction determination result for vertical writing is determined. It is output to the feature conversion means ↑5 and the rotation direction determination means 23.

When the character direction determination result outputted from the character direction determination means 20 is horizontal writing, the feature conversion means Kei 5 converts each of the characteristic matrices (f IH
1]-1-1MXN> and each feature matrix by 180°
The rotated feature matrix (f21.i-111-1l) is output to the identification means 22.Furthermore, if the character direction determination result output from the character direction determination means 20 is vertical writing,
The feature conversion means 15 first replaces the vertical feature matrix and the horizontal feature matrix with respect to the feature matrix output from the feature matrix extraction means J4, and
Swap the right diagonal feature matrix and the left diagonal feature matrix. Next, a feature matrix (f 1;, 1-1
9MXN > and the feature matrix rotated by 270° (
f 2 H,]-:l-MXN) is output to the identification means 22.

Of the two types of feature matrices outputted from the feature conversion means 15, the identification means 22 first selects the feature matrix f11 of the work type and the matching dictionary mask g stored in the dictionary storage means 214. (5> Distance defined by formula (
Similarity>Dlll is applied, and the category name C1 and distance D1 of the dictionary mask for which this distance Dm is the minimum value are output to the rotation direction determining means 23.

Similarly, the identification means 22 calculates the feature matrix f21 of the type of work and the matching dictionary mask g stored in the dictionary storage means 2]-, and calculates the distance Dl calculated by equation (5).
Category name C2 and distance D of the dictionary mask that minimizes ll
2 is output to the rotation direction determining means 23.

The rotational direction determining means 23 has a storage means such as a buffer, stores the two category names C1, C2 and the distances Di, C2 output from the identifying means 22, compares the distances, and calculates the distance. The smaller category name is output as a character name, and the rotation direction of the feature matrix from which the category name was obtained is output as the character direction.

In other words, the category name C2 is 18
These are the results obtained from the feature matrix rotated by 0°. After determining whether the category name is C1 or C2, the rotation direction determination means 23 determines whether the character is 0°, 90°, or ↓It is determined whether the direction is 80° or 270°, and the determination result is output as the character direction along with the character name. These outputs are
For example, it may be displayed on a display unit such as a CRT, or printed on a printer or the like.

As described above, this embodiment has the following advantages.

In this embodiment, the sub-pattern extracting means 13 extracts vertical sub-patterns and horizontal sub-patterns for the input pattern, and based on the extraction results, the character direction determining means 20 determines the character direction of the input pattern, whether it is vertical writing or horizontal writing. do. Based on this character direction determination result, the feature conversion means 15
Now, two different features are extracted from the input pattern, and the extracted two types of feature matrices are provided to the identification means 22. The identification means 22 performs identification processing by comparing the extracted two types of feature 4 matrices with the matching dictionary mask. The rotation direction determining means 23 determines whether or not the character is rotated by 0° and ↑80° with respect to the character direction determination result output from the character direction determining means 20, based on the identification result G from the identifying means 22, Determine the character direction of the input pattern.

As a result, a character recognition device that is fast, compact, and easy to operate can be realized. That is, in this embodiment, the character direction determining means 20 determines the line width Wv of the vertical sub-pattern and the line width Wh of the horizontal sub-pattern with respect to the input pattern. The character direction (vertical or horizontal writing) is determined based on the ratio of .

As a result, out of the four directions of 0°, 90°, 180°, and 270°, the angle is limited to two directions that are different from each other by 180°.

Further, the feature converting means] 5 extracts two different features, and the extracted features are identified by the identifying means 22, and then the rotation direction determining means 2B4 compares the identification results limited to the two directions. The character direction is determined upward from two directions that are 180° different from each other. Therefore,
There is no need to prepare matching dictionary masks in four directions of 0°, 90°, 80°, and 270° as in the conventional 5 6 , and character recognition by the identification means 22 is performed twice compared to the conventional 4 times. Since only a small amount of processing is required, processing time is reduced, and a high-speed and compact character recognition device can be realized.

Furthermore, in this embodiment, it is possible to recognize documents other than business cards as in the past, so there is no need for the operator to specify the character direction for each character or for each document, resulting in versatility and ease of use. A good character recognition device can be realized.

Note that the present invention is not limited to the illustrated embodiment, and various modifications are possible. Examples of such modifications include the following.

(i) In the above embodiment, the sub-pattern extracting means 1-3 extracts a sub-pattern from the input pattern, the feature matrix extracted from the sub-pattern by the feature matrix extracting means 14 and the feature converting means 15 is converted, and the rotation direction determining means Although the features for determining the rotation direction for 23 are generated, the features may be extracted using other methods. For example, a pattern conversion means is provided on the output side of the pattern storage means 2, the output side of the pattern conversion means is connected to the character frame detection means 1-1 and the sub-pattern extraction means 3, and the output from the character direction determination means 20 is Even if feature extraction is carried out by rotating the input pattern by a predetermined angle by the pattern converting means based on the character direction determination result and then giving the rotation result to the sub-pattern extracting means 13, the result is almost the same as in the above embodiment. Benefits can be obtained.

(ii) The Jz notation premises line width calculation means 3, the feature extraction means 10, the character direction determination means 20, the identification means 22, the rotation direction determination means 23, etc. are configured by individual circuits,
It may also be executed by computer program control or the like.

(Effects of the Invention) As described above in detail, according to the present invention, the sub-pattern extracting means extracts the vertical sub-pattern and the horizontal sub-pattern of the input pattern, and based on the extraction results, the character direction determining means extracts the vertical sub-pattern and the horizontal sub-pattern. , for example, the character direction is determined whether it is vertical or horizontal writing, so for example, if the character direction is 0°, 90°, 18
Out of the four directions of 0° and 270°, character directions in two directions that are 180° different from each other can be accurately determined.

Then, the conversion means extracts two different features from the input pattern, and the extracted features are identified by the identification means. Based on the respective results obtained, the rotation direction determination means outputs the extracted features from the character direction determination means. Since the character direction of the input pattern is determined by determining the rotation angle for the character direction determination result, the two different features extracted by the identification means can be extracted from two directions by the rotation direction determination means. The character direction of the direction is determined.

Therefore, as in the past, for example, 0'', 90.

There is no need to prepare a dictionary for checking in four directions of 180° and 270°, and the storage capacity of the dictionary storage means can be reduced, thereby making it possible to downsize the recognition device. Moreover, since the verification by the identification means is performed only twice, compared to the conventional four times, the processing time is reduced and the processing speed can be increased. Furthermore, since it is possible to determine the character direction and accurately recognize characters based on the determination results for documents other than business cards, operators can , it is not necessary to specify the character direction for each form, and a character recognition device with good operability and versatility can be realized.

[Brief explanation of drawings]

FIG. 1 is a functional block diagram of a character recognition device showing an embodiment of the present invention, and FIG. 2 (a-↓>, (a-2) to (d.-1>)
, (d-2> is a diagram showing an example of a printed document, FIG. 3(a).
~(f> is a diagram showing the vertical sub-pattern and horizontal sub-pattern of the Mincho font, and Figures 4 (a) to (e) are diagrams for explaining the binary pattern and each sub-pattern. l...・Photoelectric conversion means, 2...Pattern storage means, 3...Line width calculation means, 10...
・Feature extraction means, 11-...Character frame detection means, 12...Character frame division determination means, 13...
...Sub-pattern extraction means, 14...Feature matrix extraction means, 15...Feature conversion means, two 0 20 - Character direction determination means, 1- - Dictionary storage child stage, 2 - Identification Means: 3. Rotation direction determiner stage.

Claims (1)

[Scope of Claims] Pattern storage means for storing a read input pattern, feature extraction means for extracting features of the input pattern from the input pattern stored in the pattern storage means, and dictionary storage means for storing a dictionary. and identification means for identifying the input pattern by comparing the features extracted by the feature extraction means with the dictionary, wherein a plurality of input patterns stored in the pattern storage means are identified. a sub-pattern extracting means for scanning in the direction of and extracting a vertical sub-pattern and a horizontal sub-pattern, and calculating the line width of each of the vertical sub-patterns and horizontal sub-patterns extracted by the sub-pattern extracting means, and calculating the ratio of both line widths and a predetermined line. Character direction determining means for determining the character direction by comparing the character direction with a width ratio reference value, and character conversion or pattern conversion based on the output of the character direction determining means, thereby converting the input pattern stored in the pattern storage means. converting means for extracting features in two different directions from the character and identifying the extraction results by the identifying means; and based on the output of the identifying means, determining whether or not the output of the character direction determining means is rotated by a predetermined angle. A character recognition device comprising: rotation direction determining means for determining the rotation direction of the input pattern.