JPH0452886A - Character recognizing device - Google Patents

Abstract

PURPOSE:To correctly segment characters by providing the character recognizing device with a reading conversion means, a character segmenting means, a character recognizing means, a character element detecting means, and a separating means. CONSTITUTION:The character segmenting part 3 receives a binary image signal outputted from a binary image signal generating part 12, segments a character and generates a signal indicating the pattern of the segmented character. A segmentation control part 31 has X and Y address counters, a binary image memory 32 stores a binary image signal received by the control part 31 and outputs the stored signal at the time of segmentation. The position of each picture element is expressed by x and y coordinate values, and at the time of detecting block information, the output of a counter in the control part 31 is received as an address and picture element data in a specified memory area are outputted. At the time of receiving a signal from the memory 32, a projection part 33 segments a character by projection method. A contour tracing part 34 refers the signal outputted from the memory 32, segments a contour tracing method character and outputs a character string. A block information detecting part 37 receives the signal from the memory 32, finds out the block information, stores the information in the memory 32 and the data stored in the memory 32 are utilized by an ij detection part 38 and a longitudinal character detection part 40.

Description

Detailed Description of the Invention [Industrial Application Field] The present invention optically reads a string of characters on a manuscript, particularly a string of English letters (alphabet), that is, an English word in an English document. The present invention relates to a character recognition device that forms a value image and performs character recognition based on this, and particularly relates to improvements in character extraction.

[Conventional technology]

This type of character recognition device is used, for example, in electronic dictionaries.

FIG. 2 is a block diagram showing an electronic dictionary equipped with a conventional character recognition device. This electronic dictionary reads English words in printed English documents and displays corresponding information such as translations.

In order to translate English words on a document P using the electronic dictionary shown in the figure, the optical reading section 11 scans the document and generates an analog image signal corresponding to the read image.

The binary image signal generating section 12 receives this analog image signal and generates a binary image signal. Binary image signal generator 12
The A/D converter 12a and the multivalued image memories 12b and 2
It has a value converting circuit 12c.

The A/D converter 12a converts the analog image signal from the optical reader 11 into a digital image signal.

The multilevel image memory 12b stores digital image signals. The binarization circuit 12c converts the multi-value image signal stored in the multi-value image memory 12b into a binary image signal.

The character cutting section 13 receives the binary image signal and cuts out the characters. Character extraction refers to detecting the breaks, or separation positions, between characters in a string of characters represented by image information, and outputting the image data between the separation positions as image data for each character. .

The character cutting section 13 includes a binary image memory 14 , a character cutting control section 15 , a projection section 16 , and a contour tracking section 17 .

The binary image memory 14 stores binary image data output from the binary image signal generating section 12.

An example of a character string image stored in the binary image memory 14 is shown in FIG.

The projection unit 16 cuts out characters by a projection method based on the binary image data stored in the binary image memory 14.

FIG. 4(a) shows cutting out of characters by the projection method. According to this method, in the binary image stored in the binary image memory 14, black pixels are counted for each column in the vertical direction (direction perpendicular to the direction of the character string extending in the horizontal direction). This is called vertical projection. The lower graph in FIG. 4(a) shows the results of this projection, with the horizontal axis representing the position of each column and the vertical axis representing the number of black pixels in each column.

Then, areas A1 and A where the number of black pixels is less than or equal to the threshold Th
2. Determine A3, A4, and A5 as spaces between characters,
The characters are separated at this position.

The contour tracking unit 17 uses the 2 images stored in the binary image memory 14.
Based on the value image data, characters are extracted using the contour tracking method.

FIG. 4(b) shows character extraction using the contour tracking method. According to this method, a portion surrounded by a loop L obtained by tracing the outline of a pattern made of black pixels of binary image data is determined to be pixel data of each character. This method is also effective in cases such as h and y in "hy" where the black pixels of a character are intruding into each other in the left and right direction, so that the projection method cannot correctly cut out the characters.

The character pattern obtained as a result of cutting out is processed by the recognition control unit 18.
sent to.

The recognition control unit 11 performs character recognition by comparing this character pattern with standard character patterns stored in the recognition dictionary 19. - The recognized characters are sent to the search control section 20. Then, the word is extracted as a word, and the corresponding information is obtained by searching the meaning dictionary 21 and displayed on the display section 22.

[Problem to be solved by the invention]

However, the conventional character recognition device described above has the following problems.

FIG. 5 is a diagram showing conventional problems.

As shown in figure (a), the characters on the manuscript P are Sheriff (
The number of black pixels in each column counted in the projection method is If the loop obtained by tracing the contour in the contour tracking method surrounds two or more characters without becoming smaller than the threshold Th, the connected characters cannot be separated from each other and the segmentation is performed correctly. The problem was that I couldn't do it.

When adjacent characters are connected at the bottom end, as shown in Fig. 5(b), it is possible to try to separate them at the connected parts ■ and ■ and detect a horizontal line segment at the bottom end of the characters. Alternatively, a character that has a horizontal line segment at the bottom, such as the character ruJ shown in FIG. 4C, may be erroneously separated.

Therefore, it is an object of the present invention to solve the problems of the prior art described above, and its purpose is to be able to correctly cut out characters even when adjacent characters are connected to each other at their lower ends. An object of the present invention is to provide a character recognition device.

[Means to solve the problem]

The character recognition device according to the present invention includes a reading conversion means for optically reading a string of characters arranged horizontally on a document and converting it into binary image data representing the string of characters; receiving the image signal; Character cutting means for detecting separation positions between characters in the string of characters and outputting image data between the separation positions as a cutting result, and character recognition means for recognizing characters in response to the cutting result. , a character element detection means for detecting a character element with a predetermined characteristic by the character cutting means; and a character element detection means for detecting a character element having a predetermined characteristic; The present invention is characterized by having a separating means for detecting a position where there is an extended line segment and no character element elsewhere, and recognizing that position as the separation position.

An example of a character element having the above-mentioned predetermined characteristics is "i"
There are black dots at the top of "j" and vertical line segments such as "1" and rkJ.

[Effect]

In the present invention, character elements having predetermined characteristics, such as a black dot at the top of Fil and "j" or a vertically long line segment of Ill or rkJ, are detected, and within a predetermined horizontal distance from the detected character element. For example, within a range of about 1/2 of the average character width, a position where there is a line segment extending in the horizontal direction at the bottom edge and there are no character elements elsewhere is recognized as the separation position. There is.

Between characters whose bottom edges are connected by a series of lines, there is a line segment that extends horizontally at the bottom edge within a predetermined lateral distance from the character element, for example, within a range of about 172 points of the average character width, and otherwise Since there is a position in which there is no character element, this position is determined to be a separation position, and separation will definitely be performed.

〔Example〕

Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

(Concept) First, the concept for cutting out used in the apparatus of the present invention will be explained. In the embodiment described in detail below, when the cutout result by the projection method or the contour tracking method includes two or more characters connected to each other, the cutout (re-cutout) unique to the present invention is performed. If the cutout result contains two or more characters, it means that the width of the cutout result is significantly larger than the average character width (for example, 1.5 times or more), or that character recognition is impossible. Although it was possible, the decision is made based on the fact that the corresponding word does not exist in the dictionary as a result of the word search.

The present invention provides character elements having predetermined characteristics, such as "i", "
The independent black dots at the top of the letters in "j", rbJ, rd
J, rhJ, rkJ, rpJ, “1”, rBJ, rD
J, rEJ, rFJ, rHJ, rlJ, rKJ, rLJ
, rMJ, "N", rPJ, rRJ, and rTJ, when a concatenation occurs before and after a character having a vertical line segment, it is possible to separate the character string at such a concatenation part.

Below, we describe the algorithm for detecting sunspots in ``4'', rN, ``
The algorithms for determining the separation positions before and after "i" and "j", the algorithm for detecting vertical line segments, and the algorithm for determining the separation positions before and after characters having vertical line segments will be explained in order.

In the following explanation, each pixel, the vertex of the image,
The positions of the end points and the like are represented by X coordinate values in the horizontal direction and by X coordinate values in the vertical direction. It is assumed that the X coordinate value takes a larger value as it goes to the right, and the y coordinate value takes a larger value as it goes down.

<Algorithm for detecting black points of “i” and “j”> No. 6
The figure shows the concept of detecting the letters ``4'' and ``j.''
A case is shown in which the characters of the character string h i nJ are connected to each other by the lower syllabary. The letter “i”, “
j'' has a black dot in the upper half of the character that is separated from the rest of the character, so when contour tracking is performed, the black dot loop L
1 and the loop L2 of the other part of the character are obtained separately. This is one of the conditions (1a) for recognizing the existence of "i" and "j".

Furthermore, in order to distinguish between cases where there are loops separated due to dirt on the original P, etc., in addition to the presence of separated loops as described above, the center of at least one loop is between the left and right ends of the other loop. riJ that it is located in
Another condition (1b) for rjJ detection is assumed.

That is, the coordinate values of the top, bottom, left and right ends of the upper loop L1 are (Yx
u, YlL, XIL, XIR), and the coordinate values of the top, bottom, left and right ends of the lower loop L2 are (Y2U, Y2L,
X2Lr , X2C are determined by the following formulas.

XIC-(XIL+XIR)/2 X2C= (X2L+X2R)/2, and X2L≦X. ≦X2R or X1L≦X2c≦X1. l: Check whether condition (1b) is satisfied. and the above condition (1a)
, (1b) are both satisfied, it is determined that riJ, "j" exists.

Note that the determination as to which of the one loop and the other loop is the upper loop, that is, the loop surrounding the black dot, is made by comparing the lower end positions of the respective loops. That is, the one located above the lower end, in other words, the one whose y-coordinate value YIL%Y2L of the lower end is smaller, is regarded as the black circle of the characters "i" and "j".

<Algorithm for determining separation positions before and after "i" and "j"> When the presence of FiJ, rj, and J is detected as described above,
Next, the separation positions before and after Fil and "j" are determined. This is based on the following algorithm.

FIG. 7 is a diagram showing an algorithm for determining separation positions before and after "i" and "j". As shown in the figure, within the range w/2 (w is the average character) from the center of the black point to the left and right, we examine the rows of pixels in the vertical direction, and find out if there is one block (a collection of consecutive black pixels). And the number of black pixels is less than or equal to a predetermined vertical range, that is, a predetermined height Ht from the bottom edge of the character area (for example, Ht=0,2XH (H is the average character height expressed in number of pixels)) The column where is recognized as the separation position. If there are a plurality of columns that satisfy such a condition (condition (2)), the one located in the center of them is recognized as the separation position C.

If there is no column that satisfies the above condition (2), it means that no suitable separation position has been found, and the separation position is not recognized.

Also, if there is an inner column (column that does not contain any black pixels) within the range up to w/2, it is already known that that column is the separation position (by the projection method or contour tracking method) Therefore, the separation position will not be certified again.

<Algorithm for detecting vertically long line segments> Next, an algorithm for detecting vertically long characters will be explained with reference to FIG. 5(a). In the illustrated example, the read ralre
The ralrJ parts of adyJ are connected to each other.

First, the heights of characters cut out by the projection method, contour tracking method, etc. (or a group of characters obtained as a cutout result due to incomplete separation) are expressed as Hl, H2, H3...H
m (m is the number of cut-out results), and the maximum value HI1mM and the minimum value H11mM of H□ to H1. ．． Check whether the following relationship holds between (H, , , / Hffi, ,)≧a ・=−condition (
3) (here α is greater than 1, for example 1.3).

) Next, check whether the following judgment condition (4) is satisfied. That is, the number of blocks is 1 and the length is a predetermined value β
When a predetermined number n (for example, 1) or more of consecutive columns with XHmax (Hmax is the maximum character height in the word, β is 0.9, for example) or more are continuous, the existence of a vertical line segment is recognized.

FIG. 8(a) shows that the character "1" has a vertically long line segment, that is, it satisfies the above-mentioned determination condition. As shown in the figure, in the vertical line segment LS, a predetermined number or more (4 in the illustrated example) of blocks BR having one block and a length of βXHmax or more are consecutive.

If condition (3) or condition (4) is not satisfied, it is determined that the vertical line segment does not exist.

<Algorithm for determining the separation position before and after a character with a vertically long line segment> As described above, when the presence of a vertically long line segment is detected in the re-extraction target, the character having the vertically long line segment and the characters before and after it are separated. Separate the The separation position is determined as follows.

That is, as shown in FIG. 8(b), from the left end of the vertical line segment to the left, 7w (w is the average character, γ is, for example, 0.4
), and from the right end of the vertical line segment to the right,
It is checked whether there is a column that satisfies the following separation condition (5) in each range up to γW.

That is, the number of blocks is 1, the number of sunspots of the block is less than or equal to δXHmax (δ is, for example, 0.2), and the upper end of the block is ε×Hmax from the lower end of the column (ε is, for example, 0.2). It is lower.

In FIG. 8(C), the above condition (5) is satisfied for the column of coordinate values X+, so the separation condition is satisfied. Therefore, it is separated at this position. On the other hand, in the column of coordinate values Xk shown in FIG. 3D, the number of blocks in the column is 2, and the above separation condition is not satisfied.

Furthermore, although the number of blocks in the column of coordinate value X1 is 1, the position thereof is not within the range of εXHmax from the bottom end of the column, so the above separation condition is not satisfied.

When the above separation conditions are satisfied, the separation position is determined as follows.

That is, as shown in FIG. 9, the widthwise center of the vertical line segment (
If there is a column that satisfies the above condition (5) within the range w/2 left and right (center) of the plurality of columns constituting a vertical line segment, the lines are separated by that column. When there are several consecutive such columns,
That is, when there are a plurality of consecutive columns satisfying such conditions, the center position is recognized as the separation position C.

Note that when a column that satisfies condition (5) is separated by a column that does not satisfy condition (5), it is separated at the position of the column or group of columns that satisfies it, which is closer to the vertical line segment.

Furthermore, if there is an inner column within the range up to w/2,
Since it is already known (by the projection method or contour tracking method) that the column is the separation position, the separation position is not determined again.

(Configuration of Apparatus) FIG. 1 is a block diagram showing an example of an electronic dictionary incorporating a character recognition device equipped with a character extraction section that performs extraction according to the above concept. This electronic dictionary optically reads English words on the document P and displays the corresponding information.

In the figure, the same components as in the conventional example are given the same reference numerals, and the explanation of the structure will be omitted.

Reference numeral 3 denotes a character cutting section which is a feature of the present invention, which receives the binary image signal output from the binary image signal generator 12, cuts out characters, and generates a signal representing the pattern of the cut out characters.

FIG. 10 is a block diagram showing the configuration of the character cutting section 3. As shown in FIG.

In the figure, reference numeral 31 denotes a character cutting control section, which controls character cutting based on the binary image signal received from the binary image signal generating section 12.

As shown in FIG. 11(a), the cutting control section 31
Address counter 31-1, Y address counter 31-
It has 2.

A binary image memory 32 stores the binary image signal received by the character cutting control section 31 and outputs it when cutting out. This memory is called a bitmap memory or brain memory, and stores the data of each pixel in a memory area associated with the position of each pixel. That is, the position of each pixel is represented by an X coordinate value, a y coordinate value, and this X coordinate value,
The data of that pixel is stored in the memory area specified by the X address and the X address representing the y-coordinate value.

When detecting block information as described later, the outputs of the X address counter 31-1 and Y address counter 31-2 of the extraction control unit 31 are received as addresses, and the pixel data of the memory area specified by these addresses is output. do.

Reference numeral 33 denotes a projection unit that displays the 2 output from the binary image memory 32.
In response to the value image signal, character segmentation is performed using the projection method described in the prior art section, and the segmentation results are output.

Reference numeral 34 denotes a contour tracking section which performs character segmentation using the contour tracking method described in the prior art section by referring to the binary image signal from the binary image memory 32, and outputs a string of characters obtained as a result of the segmentation. .

37 is a block information detection unit that receives the binary image signal from the binary image memory 32, and detects the block information (X coordinate value indicating the column in which the binary image signal is scanned in the vertical direction and continuous black dots obtained by scanning). (i.e., the number of blocks, the y-coordinate value indicating the top end of the block, and the y-coordinate value indicating the bottom end of the block) are stored in a part of the memory in correspondence with each other, and the ij detection section 38 and the vertically long character detection section Available for use at 40.

FIG. 11(a) is a diagram showing the internal configuration of the block information detection section 37, as well as a part of the extraction control section 31 and the binary image memory 32.

The extraction control unit 31 first selects the X address counter 31-
1 sequentially selects each column of the image stored in the binary image memory 32, and the Y address counter 31-2 sequentially selects each pixel in each column.

That is, the image is sequentially scanned in the vertical direction (Y address direction). This includes an X address counter 31-1 and a Y address counter 31-2 that give addresses to the binary image memory 32.
This is done by counting up the contents in a predetermined order.

The pixel data is, for example, as shown in the upper half of FIG. The pixel data is, for example, as shown in the lower half of FIG. 11(b).

The output of the Y address counter 31-2 is supplied to the binary image memory 32, as well as to a block upper end address register 37-8 and a block lower end address register 37-9, and is latched at a timing described later. Similarly,
The output of the X address counter 31-1 is supplied to the binary image memory 32 and also to the X address register 37-10.
It is also supplied to the signal and latched at the timing described later.

The block end detection unit 37-2 receives the pixel data output from the binary image memory 32, and detects the upper end of the block (YUII, YUI2, YU13 in column X1 in FIG. 11(b)).
, YU21 in column X2) and the bottom end (YLII, YL12, YL13 in column X1 in FIG. 11(b), YL in column X2)
21) is detected. In this detection, the pixel data of each column is output in order from the top of each column downward, and the point of change from a white pixel to a black pixel is recognized as the top of the block, and the change from a black pixel to a white pixel is detected. This is done by identifying a point as the bottom edge of the block. End detection section 37- having such a function
2 can be constituted by, for example, flip-flops 37-3, 37-4 and AND gates 37-5, 37-6 connected as shown in FIG. 11(C).

The block upper end address register 37-8 is the upper end detection signal 37-2a output from the block end detection section 37-2.
is received, and the output of the Y address counter 31-2 at this time is latched. The latched value is the coordinate value of the top end of the block.

9 to the block lower end address register 37 is the lower end detection signal 37-2 output from the block end detection section 37-2.
b, and latches the output of the Y address counter 31-2 at this time. The latched value is the coordinate value of the bottom end of the block.

The block number counter 37-7 is connected to the block end detection section 3
receiving the upper end detection signal 37-2a output from 7-2;
Count this. This count value represents the number of blocks.

When the binary image signal of each column has been scanned from the top end to the bottom end in this way, the X coordinate value of that column is stored in the X address register 37-10 and in the block number or block number counter 37-7. , the coordinate values of the upper end of the block are stored in the block upper end address register 378, and the coordinate values of the lower end of the block are stored in the block lower end address register 37-9.

Block information memory 37-11 includes X address register 37-10. , the block number counter 37-7, the block upper end address register 37-8, and the block information output from the block lower end address register 37-9, respectively, are stored in correspondence with each other.

The above is the block information detection method in the block information detection section 37, and the block information obtained in this way is sent to the ij extraction section 35.
and is sent to the vertically long character cutting section 36.

The ij cutout section 35 is connected to the ij detection section 38 and the ij bone separation section 3.
It has 9. On the other hand, the vertically long character cutting section 36 includes a vertically long character detecting section 40 and a vertically long character separating section 41 .

The ij detection unit 38 refers to the binary image signal from the binary image memory 32 and detects "i" and "j" as described above.

ij bone separation 39 is performed when the letters “i” and “j” are included in the re-excision target.
When it is determined that "i" and "j" are included, the block information is referred to and separation is performed as described above before and after the "i" and "j".

The vertically long character detection section 40 refers to the binary image signal from the binary image memory 32 and detects a vertically long character as described above.

When a vertically long character is detected, the vertically long character separation section 41 refers to the block information and separates the characters before and after the vertically long character as described above.

The operations and operations of the electronic dictionary incorporating the character recognition device according to this embodiment are as follows.

First, a string of characters on the original P is read by the optical reading section 11 and output as an analog image signal. Then, in the binary image signal generation section 12, an A/D converter 12a
1 multivalued image memory 12b 1 through binary conversion circuit 12c, 2
converted into a value image signal. The binary image signal output from the binary image signal generating section 12 is received by the character cutting control section 31 of the character cutting section 3, stored in the binary image memory 32, and used for character cutting.

Character cutting is performed under the control of the cutting control section 31.
Projection unit 33, contour tracking unit 34, block information detection unit 37
, ij cutting section 35, and vertically long character cutting section 36.

FIG. 12 is a flowchart showing the character extraction operation.

The extraction procedure will be explained below according to this flowchart.

As shown in the same figure and FIG. Characters are cut out according to the method (FIG. 4(b)) (Sl).

Next, it is determined whether re-cutting is necessary for the cropping results obtained by the projection method and the contour tracking method described above (S2).

In this embodiment, this determination is made by calculating the average character size based on the above cutout results, and then calculating the average character width by multiplying the average character by a predetermined coefficient, for example, 1.5. This is done by determining whether or not there is something (obtainable).

When it is determined in S2 that re-extraction is necessary, block information is detected in the block information detection section 37 (S3).

Next, the above conditions (1a) and (1
b) is satisfied, that is, if there is a character “i” in the re-extraction target
”, “j” is checked (S4).

When it is determined in step S4 that there is, i, j
Check whether separation condition (2) is satisfied (
S5). If it is determined in step S5 that the separation condition is satisfied, the material is separated and cut out at the separation position determined as described above (S6).

When the separation is completed in step S6, in step S4,
If "i", "j" are not detected, and step S
If separation condition (2) is not satisfied in step S7, the process proceeds to step S7, where the vertically long character detection section 40 checks whether there is a vertically long character among the objects to be recut. This depends on whether the above conditions (3) and (4) are satisfied.

In step S7, if it is determined that there is a vertically long character, the detection result is sent to the vertically long character separation unit 41, and based on the block information sent from the block information detection unit 37, separation condition (6) is satisfied. It is determined whether the
).

If satisfied, separate at an appropriate separation position (
S9).

Then, the re-cutting result obtained in this way is sent to the recognition control unit 18, where character recognition is performed (SIO).

Further, when a vertically long character is not detected in step S7, and when the separation condition is not satisfied in step S8, the process also proceeds to step S10 and character recognition is performed.

After character recognition, judge whether it was recognizable or not.5
ll), exit if possible.

If this is not possible, handle the error.

If it is determined in step S2 that re-cutting is not necessary, character recognition is immediately performed (S12).

Then, it is determined whether the recognition is possible or not, and if it is possible, the process ends. On the other hand, if it is not possible, the process advances to step S3 and re-cutting is performed.

After character recognition is performed as described above, the recognition result is sent to the search control section 20, the meaning of the word is checked with reference to the meaning dictionary 21, and the result is displayed on the display section 22.

In the above embodiment, when a wide character is detected in step S2, re-cutting is immediately performed.
If the word cannot be found in the meaning dictionary 21 even after searching with 0, this may be communicated to the character extraction unit 3 to perform re-extraction.

〔Effect of the invention〕

As explained above, according to the present invention, character elements having predetermined characteristics, such as black dots at the top of "i", "rj", rbJ, "dco, rhJ, rpJ, rlJ, rBJ,
rDJ, rEJ, rFJ, “H” rIJ, rKJ, rL
Detects the vertical line segments of J, "M", rNJ, rPJ, rRJ, rTJ, and separates them at a portion that has a horizontal line segment at the bottom end within a predetermined range before and after it and does not have any other character elements. Since we have done this, even if characters are connected to each other by a sequence, etc., we can definitely separate them. Therefore, the success rate of character extraction increases.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of this embodiment, Fig. 2 is a block diagram of a conventional example, Fig. 3 is a diagram showing an example of storage in the binary image memory of Fig. 2, and Fig. 4 (a) is a projection method. Figure 4(b) is a diagram showing the extraction method using the contour tracking method, Figures 5(a) to (C) are diagrams showing the problems of the conventional method, and Figure 6 is the present example. 7 is a diagram showing an algorithm for determining the separation position of the character Fil in this embodiment. FIGS. 8(a) to (d) are vertically long characters in this embodiment. FIG. 9 is a diagram showing an algorithm for determining the separation position of vertically long characters in this embodiment. FIG. 10 is a block diagram of the character cutting section in FIG. 1.
Figure 11(a) is a diagram showing the internal configuration of the block information detection section in Figure 1, Figure 11(b) is a diagram showing the concept of block information detection in Figure 1, and Figure 11(C) is the diagram shown in Figure 11. FIG. 12 is a flowchart showing the character cutting operation of this embodiment. 3...Character extraction unit, 31...Character extraction control unit, 31-1...X address counter, 31-2...
... Y address counter, 32 ... Binary image memory, 33 ... Projection section, 34 ... Contour tracking section, 35, ij cutting section, 36, Vertical Character cutting section, 37...Block information detection section, 37-2...
...Block end detection unit, 37-7...Block number counter, 37-8...Block upper end address register, 37-9...Block lower end address register, 37-10...X address register, 37-11...Block information memory,
38...ijj output section, 39...ij decomposition separation section 40...vertical character detection section, 41...vertical character separation section. 2 ζ” In the lawsuit, Meyori → O 1 ≧ 6 η and 9.1 Dari goes down 2 3rd figure Dawn crashes IF5 Figure (b) (a) (b) (c) Is there a small sound? Lugo 11 Sushosanhe-di-) Descending Figure 6) Kubu% (ni", 1 t of bubble filling) 3 Shiji Risum Tsukiji Figure 7

Claims (3)

[Claims] (1) A reading conversion means that optically reads a string of characters arranged horizontally on a document and converts it into binary image data representing the string of characters; The character cutting means includes: a character cutting means for detecting separation positions between characters and outputting image data between the separation positions as a cutting result; and a character recognition means for recognizing characters in response to the cutting result. a character element detection means for detecting a character element having predetermined characteristics; and a line segment extending in the horizontal direction at the lower end within a predetermined lateral distance from the character element detected by the character element detection means. 1. A character recognition device comprising: a separating means for detecting a position where a character element is present and where there is no character element elsewhere, and recognizing that position as the separation position. (2) The apparatus according to claim 1, wherein the character element detection means detects black dots above "i" and "j" as character elements having the predetermined characteristics. (3) The apparatus according to claim 1, wherein the character element detection means detects a vertical line segment as the character element having the predetermined characteristics.