JPS61196382A - Character segmenting system - Google Patents

Abstract

PURPOSE:To execute character segmenting with a simple constitution and high accuracy by performing column scanning in the direction of respective opposite sides from top to bottom sides of circumscribed frames of character patterns, classifying and then based on the results, carrying out row scanning in the circumscribed frame. CONSTITUTION:Quantized character pattern columns are scanned in column direction and the width of black spot histogram is detected, which is compared with the average character width to determine the area wherein character segmenting process is executed. The area is row scanned and black spot histogram in row direction is prepared and based on the column and row direction black spot histograms, the character circumscribed frame is determined. The frame is row scanned and the variation points where classification change amongst the four different classes are detected, the variation points successively stored, compared with combination of prescribed variation points in order, and coinciding variation points are detected. According to the coordinates of coinciding variation points, the character segmenting position is determined.

Description

[Detailed Description of the Invention] (Industrial Application Field) The present invention relates to a character extraction method, and more specifically, reads characters written on a form, and creates a character pattern string based on the read characters for each character area. This paper relates to a character extraction method for separating and extracting characters.

(Conventional Technique #i) An optical character recognition device (hereinafter abbreviated as OCR) scans characters written on a form line by line, converts the optical signal into an image signal using a photoelectric converter, and converts it to a line. , stored in file. The line buffer is sequentially read out, the character pattern string is separated into character areas, and recognition is performed using the separated character patterns.However, the character extraction method of extracting a single character area from the character pattern string is an OCR method. It greatly affects performance.

Next, a conventional character extraction method for separating a single character area from character string pattern data stored in an OCR line buffer will be described.

In OCR, a character pattern in the line buffer is read by scanning one column from the top to the bottom of a line buffer in which one character string is stored, and sequentially moving the columns in a direction perpendicular to this scanning.

A histogram is created by counting black dots (black dots in the text area and white dots in the background area) during scanning of one line, and the area of one character is determined by referring to the black dot histogram.

FIG. 10 is a diagram showing a pattern sequence using a conventional black point histogram. In the figure, 100° 101 is a character pattern, which is a pattern stored in a line buffer in OCR. 102 is character pattern 100
, 101 in the column direction. In addition, in the figure, reading starts from a specified position at the left end of the line buffer, and while reading one column, a black point histogram is created for that column, and this histogram is compared with a threshold value α (α: constant). The column whose histogram is larger than α is the starting point, and the column whose histogram is smaller than the threshold α is the ending point, and the area from the starting point to the ending point is cut out as a region of one character.

(Problems to be Solved by the Invention) However, in the case of handwritten characters, in the case of handwritten characters, the writer writes the characters at an angle, or the characters are written outside the character writing frame, or Adjacent letters overlap due to reasons such as someone hitting a part of the letter.

There was a problem in that a character pattern of two or more characters was cut out as one character. Also, as shown in Figure 10,
Since the character patterns 100 and 101 overlap in the column direction, their black point histogram 102 is formed as one area. Furthermore, if the length from the start point to the end point of the black point histogram is determined and it is determined that there are two or more characters, even if the position corresponding to the average character width is used as the cutting point, some characters other than the relevant character may be mixed in. There was a problem that some characters were missing.

This invention is intended to solve these problems.
The purpose is to provide a highly accurate character extraction method with a simple configuration.

(Means for Solving the Problems) In order to solve the above-mentioned problems, the present invention converts a quantized character pattern string obtained by photoelectrically converting a character string written on a form into a quantized character pattern string for each character. The character extraction method for separating and extracting characters is configured by the following means.

This invention stores a quantized character pattern string in a line buffer memory, scans the line buffer memory one column corresponding to the column direction of the character string, and creates a black point histogram in the column direction. A means for sequentially detecting the width of the black point histogram in the column direction for each column, and a means for detecting the width of the black point histogram in the column direction by comparing the average character width of the specified target character with the width of the black point histogram in the column direction. A black point histogram in the row direction is obtained by determining a region for character extraction processing based on the number of characters included in the black point histogram, and scanning the character pattern string in the region row by row in the row direction. means for detecting a character circumscribing frame of a character pattern string from the black point histogram in the column and row directions; a storage means for storing the character pattern string within the character circumscribing frame; a detection means for reading out the contents of a character pattern string from a storage means as it scans toward the opposite side and detecting whether the contents are a character portion or a background portion; and a background portion detected by scanning from the upper side. A classification means for classifying a character pattern string within a character circumscribing frame into four types: a background part detected by scanning from the bottom side, a single character part, and other background parts, and a change in which the classification changes by performing line scanning again. A change point detection means for detecting points, sequentially storing them, and sequentially comparing them with a predetermined combination of change points to detect a matching change point, and means for determining a character cutting position based on the detected change points. It consists of

(Operation) According to the character extraction method having the above configuration, the operation is as follows.

The quantized character pattern string is scanned column by column to detect the width of the black point histogram in the column direction, and this is compared with the average character width to determine the area for subsequent character extraction processing. Ru. Further, the area is scanned in rows to create a black point histogram in the row direction, and a character circumscribing frame is determined from the black point histogram in the column direction and the row direction. Furthermore, the characters are scanned from the upper and lower sides of the character circumscribing frame to the opposite sides, and are classified into the above-mentioned four types of parts. Then, the lines within the character circumscribing frame are scanned again to detect change points where the classification changes, and the change points are sequentially stored and successively compared with a predetermined combination of change points to detect matching change points. the result,
A character cutting position is determined based on the coordinates of the matching change point.

(Example) Hereinafter, one embodiment of the present invention will be described based on the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention. In the figure, 200 is an image signal from a photoelectric conversion unit (not shown), 201 is a line buffer 5202, a black point histogram creation circuit 22o, and a circumscribing frame detection circuit 221.
and a character determination circuit 222. 203 is a data switching circuit, 204 is a pattern memory, and 205 and 208 are an X counter in the X direction and a Y counter in the X direction, which generate addresses for the pattern memory. 207 is a control circuit. 208 is a pattern area classification circuit, and 209 is a circuit for detecting a change point from a white point to a black point. 210 is a pattern area change point detection circuit; 211 is a cutout area detection circuit; 2
12 to 214 are registers for determining the cutout area.

The operation of this embodiment will be explained below using the block diagram shown in FIG.

The character string on the form is converted into a binary image signal 200 by a photoelectric converter and stored in a line buffer 201. The following processing is performed under the control of the control circuit 207. The control circuit 207 converts the image signal stored in the line buffer 201 by 1 from the beginning position of the line buffer 201.
The character pattern data for one line is read out sequentially, and the process ends when all the character pattern data for one line is read out. Further, the control circuit 207 starts up the black dot histogram creation circuit 220 at the same time as reading the pattern data from the line buffer 201 for one train. Sunspot histogram creation circuit 220
Now, by counting the number of sunspots that are being read in one column, a blackspot histogram for that column is created, and the process of storing 0 or more in the histogram memory 230 included in the sunspot histogram creation circuit 220 is repeated for one row and all columns. The process ends when the black point histogram is stored in the histogram memory 230.

After creating a sunspot histogram for one line, the control circuit 207 reads the histogram memory 230 in the sunspot histogram creation circuit 220 from the beginning and detects a block by referring to the blackspot histogram. From the histogram memory 230 inside,
The black point histograms are sequentially read out, and the black point histograms are compared with a threshold value α (α: constant, however, α=1 in this example). If the histogram is large, it is considered as a starting point candidate for a block of characters, and Advance the storage address, count the columns in which the read black point histogram is greater than the threshold value α, and if the β (β: constant, however, β = 2) columns are continuous, select the starting point candidate. Use as starting point. The row continues to advance, and after the starting point is detected, the row whose black point histogram becomes smaller than the threshold α for the first time is set as the end point, and the area indicated by the length from the starting point to the ending point is set as a block. Next, the control circuit 207 activates the character determination circuit 222 and sets the length of the detected block to threshold values γ1 and γ2 (γ1 and γ2 are constants, where 1 In this embodiment, γ=75.γ2=125). Then, the length W of the block is a threshold value γ! If it is smaller than γ!, the block is determined to be one character, and γ! ≦W≦
When γ2, it is determined that there are two characters, and when W>γ2, it is determined that there are three or more characters. Further, after determining the block, the control circuit 207 activates the circumscribing frame detection circuit 221 for the block to detect the circumscribing frame. Furthermore, when the circumscribing frame of this block is detected, the character pattern within the circumscribing frame is transferred to the pattern memory 204. here,
In the case of W>γ2, that is, if the block is determined to be 3 or more characters, perform the cutting process from the starting point to γ2, divide the first and second characters, and then use the resulting cutting point as the starting point to further divide the second character. Sequential extraction is performed, such as dividing the third character, and processing up to W is performed. Furthermore, if W>γ2, the area from the starting point to γ2 is transferred to the pattern memory 204 within the circumscribed frame, and the rest is transferred from the starting point to γ2 by the following process.
Transfer is performed again when the cutout point is determined. Here, as shown in FIG. 2, which will be described later, the left end of the upper side of the character circumscribing frame is taken as the origin, the lower side position is FB, and the right side position is PR.

Next, processing of characters determined to be two or more characters in the above character determination will be explained based on FIG. 1.

The control circuit 207 sets the X counter 205 and the X counter 206 that give the address of the pattern memory 204 to the left end position of the upper side of the circumscribed frame of the character, and sets the X counter 208
is incremented to scan toward the lower side of the character circumscribing frame, and the addresses of the pattern memory 204 are set to (x, y) for the X and Y axes, and the values of the X counter and X counter are used, respectively. The contents of the pattern memory 204 at the position indicated by the address are expressed as PM (x, y). In this embodiment, the white point is expressed as PM (x, y).
= 0, the black point is PM (X, V) = 1, the 0 point detected when scanning from the upper side is PM (x, y) = 2, the D point detected when scanning from the lower side is PM (x ,y)
= 4. Therefore, the pattern memory 204 in this embodiment has a data width of 3 bits for 1 mesh. In the pattern area classification circuit 208, an address is set at the left end of the upper side of the character circumscribing frame, and the character pattern is read out from the pattern memory 204.

When PM (x, y) = 0(7), (PM (X
, Y).

OR, 2) as a new PM (x, y) and the switching circuit 20
3 to the corresponding address in the pattern memory 204.

The control circuit 207 is a change point detection circuit 2 from a white point to a black point.
When 08 detects a black point where PM (x, y) = 1, it stops scanning the column, increments the X counter 205 by 1, and scans the next column from the upper side of the character circumscribing frame.
Also, when scanning starts from the top side of the character circumscribing frame and reaches the bottom side, the scanning of the column is aborted, and the next column is scanned.0 or more scans are repeated sequentially, and the process ends when the rightmost column of the character circumscribing frame is processed. do. When the scanning from the upper side is completed, the control circuit 207 sets the X counter to the left end of the lower side of the character circumscribing frame, scans from the lower side to the upper side, and performs the same operation as when scanning from the upper side. perform processing,
However, when PM (x, y) = 0, (PM
(x, y), OR, 4) is stored in the pattern memory 204 as PM(x, y). The control circuit 207 completes the processing of the rightmost column in the same manner as the scanning from the upper side. When the two types of scanning are completed and the patterns within the character circumscribing frame are classified, the control circuit 207 controls the X counter 205 and the X counter 206. is set at the upper left edge of the character circumscribing frame, and horizontal scanning is performed to detect the character cutting area.

Here, an example will be used to specifically illustrate the above column scanning. FIG. 2 is a diagram showing a specific example of vertical scanning in this embodiment. In the figure, 100 and 101 are character patterns, 103 is a scanning direction from the top side to the bottom side, and 10
4 shows the scanning direction from the lower side to the upper side, and FIG. 3 is a diagram showing the processing result of the column scanning in FIG. 2. In the figure, a white point detected during scanning from the top side to the bottom side (the text part is a black dot and the background part is a white dot) is designated as a point C, and a set of C points is designated as a C area. Furthermore, if a black spot is detected during the scanning, the scanning of the column is stopped at that point and the next column is processed. Here, it is assumed that area A is a collection of black points. Similar processing is performed when scanning from the lower side to the upper side, the white point detected during this scanning is defined as point D, and the set of D points is defined as area D. 2
By scanning twice, white points other than point C and point D, that is, the above-mentioned 2
The white point that was not scanned in the previous scan is defined as B point, and the set thereof is defined as B area.

Next, detection of a single character cutout area will be explained based on FIG.

First, the pattern area change point detection circuit 210 is connected to the control circuit 2.
07, character pattern data is read from the pattern memory 204 and lines are scanned within the circumscribed frame. The pattern area change point detection circuit 210 processes the character pattern data from the pattern memory 204, and while the character pattern data of the point currently being processed is being processed, the pattern area change point detection circuit 210 processes the character pattern data of the point immediately before that point. It holds pattern data and compares the currently processed character pattern data with the previous character pattern data. If it is determined that the comparison result has changed, the coordinate position of the previous point is detected and held. In other words, PM(x-1°y) and PM(x
, y), and if they are not equal, the X-axis coordinate x-1
is stored in xREG I214. In the cutout area detection circuit 211, the pattern area change point detection circuit 210
When the change point is detected, PM(X-1, y) is held in the register. In the cutout area detection circuit 211,
It has three status registers (not shown) that hold the PM (x-1, y), and the status registers have a configuration in which the contents of the registers are shifted to adjacent registers when the change point is detected. It becomes. Furthermore, when the change point is detected and the shift of the state register is completed, it is detected whether the contents of the three types of state register match the following states stored in the control circuit 207. STI status register.

If Sr1 and Sr1, 5Tl=4 and 5T2=0
and 5T3=2, or 5Tl=2 and 5T2=0 and 5T3=4, or 5T2=2 and 5T3=4, or 5T2=4 and 5T3=2. However, it is assumed that Sr1 is the content of the current coordinate position. Therefore, if the state register matches the combination,
The decision signal from the cutout area detection circuit 211 is 7REG2.
12 and xREG lI213. The contents of the X counter 205 or y counter 206 stored in each register at that time are 71? It is output as X2. Also, X! is the content of the X counter 205 stored in xREG I214 when the status register ST3 of the pattern area change point detection circuit 210 is at point C or point D. Also, if the status register matches the combination, horizontal scanning of that row is aborted, and Y
The counter is incremented and a new next row is horizontally scanned. At the next point where 0 or more horizontal scans are all completed within the circumscribed frame, xl (xREG I), X2 (xREG
II) Determine the cutting position based on Vt (yREG■).

The pattern transfer method will be described below using the example pattern shown in FIG. 4 in which the cutout position has been determined. Also,
FIG. 4 shows the patterns and cutout positions stored in the pattern memory 204 in the block diagram of FIG. 1. The coordinates are the horizontal axis as the X axis and the vertical axis as the Y axis.
It is assumed that the pattern memory 204 is located in the fourth quadrant. XM and YM indicate the size of the pattern memory 204, and in this embodiment, XM=YM=128
It was a mesh. PR and PR are pattern memory 20
The circumscribing frame of the pattern stored in 4 is represented by four straight lines: X=G, X=PR, Y=0, Y=PR. In FIG. 4, 300 and 301 are patterns, and straight lines Y=y1, It has become. In Fig. 5, when (1) is 1, it means that it was a white point during column scanning from the bottom side to the top side, and (1)
) is 0, it means that the point is other than a white point. Further, when (2) is 1, it means that the point was a white point during column scanning from the top side to the bottom side, and when (2) is 0, it means that it was a point other than a white point. Furthermore, when (3) is 1, it means a point that is a black point, and when (3) is 0, it means a point that is a white point. Therefore, the pattern data to be transferred is (3
) is the only data shown. By incrementing the Y coordinate of the mesh on the straight line represented by x=0 from the point Y;O by one, the pattern data is transferred to the point Y=PB. Example 1: Increment the X coordinate after the transfer is completed. The above transfer is repeated for each column, and when the transfer of the column X=x is completed, the transfer from the next column to the column X=x2 is Y.
For coordinates from yI to FB, pattern data is masked and a fixed value of 0 is transferred. The transfer of the pattern 300 ends when the column of X=x2 is transferred. The pattern 301 can also be transferred using a similar method. Furthermore, for data in which one character is included within the circumscribing frame, the pattern within the circumscribing frame can be transferred using a similar method.

Next, the process flow of this embodiment will be explained in detail based on the flowcharts shown in FIGS. 6, 7, and 8. Here, FIG. 6 shows the overall flow, and FIGS. 7 and 8 show flowcharts of classification of pattern areas and determination of cutout areas by two upper and lower scans, respectively. First, the overall flowchart in FIG. 6 will be explained. In 3400,
Start reading operation. At 3401, one column of pattern data stored in the line buffer is read out, and the black point histogram creation circuit 220 in FIG. 1 creates a black point histogram and stores it in the histogram memory 230. 54θ2
In step 3403, the processed characters are managed,
The following process is repeated until all characters in the line have been extracted.

At step 5404, the black point histogram is read out from the histogram processor, and the starting point and ending point of the black point histogram are detected and set as a block. Also, the length of the block and the threshold γ1
, γ2 and retains how many characters the block consists of. In 5405, the circumscribing frame of the block is detected by the circumscribing frame detection circuit 221 in FIG. 1, and the pattern data within the circumscribing frame is stored. is transferred to the pattern memory 204. At 8406, based on the judgment result of the length of the held block, if it is one character, the pattern data in the pattern memory 204 is transferred to the output stage and processing proceeds to the next character; if it is two or more characters, the following is executed. Perform processing.

In step 5407, column scanning is performed from the upper and lower sides of the circumscribing frame to the opposite sides, the pattern area is classified, and the results are stored in the pattern memory. At step 5408, the lines within the circumscribed frame are scanned, the classification results are read out from the pattern memory, and the cutout area is detected to determine the cutout position. At step 5409, the pattern in the pattern memory is transferred according to the cutting position. When all the patterns in the pattern memory have been transferred, the next character is processed.

Next, detailed flowcharts of the processes 5407 and 5408 in FIG. 6 are shown in FIGS. 7 and 8, and the operations will be explained in order.

FIG. 7 shows the flow of classifying character pattern areas and detecting points of change from white dots to black dots.

5500, when character pattern data is input, 55
01 and 5502 are initialization, and the coordinates of X and 7 in the pattern memory are set to the upper left edge of the character circumscribing frame, and the value U/D indicating the scanning direction is set to 2 because scanning is performed from the upper side to the lower side. . 5503, the contents of the pattern memory are PM (x.

y)=1 (black point), the process is moved to 5507, and PM
(x, y) When there is no (white point).

5504-11' The contents of the pattern memory are PM(x, y
) = (PM (x, y), OR, U/D).

At 5505, the value of the y counter is incremented or decremented depending on the scan direction. 850
In step 8, one row is managed and the process is continued in step 5 until the processing of one row is completed.
Return to 503 and repeat the same process, PM (x, y) =
5 when 1 is detected or one column scan is completed
The y counter is set to the scanning start point (on the upper or lower side) in 507, the X counter is incremented in 5508, and the process from 5503 is repeated until the X counter matches the right end (PR) of the character circumscribing frame. If the X counter matches the right end during scanning, the scanning direction is set from the bottom side to the top side, and the above process is repeated.

At this time, U/D is set to 4 in 5511. Similar processing is performed for the scan of U/D=4, and when all is completed, the processing shown in the flowchart of FIG. 8 is performed.

FIG. 8 shows the flow of detecting a change point in a pattern area and determining a cutout area.The cutout area is determined by horizontally scanning the character circumscribing frame of the pattern memory from the left end of the upper side. X for s eoo and s sot respectively
Initialize the counter and X counter. In the 58G2, a status register 5TI is used to hold changes in the area during row scanning.
-3T3 is initialized, Sr1 indicates the state of the current position, Sr1 indicates the area before the current area during scanning, and STI indicates the area before STZ. In 5803, the contents of the pattern memory PM(x.

y) is compared with Sr1 and if they match, proceed to 5605;
If they do not match, the coordinates are a change point, so 5e0
4, the contents of sT2.5T3(7) are STI and S, respectively.
Shift to T2. In 5805, PM (x,
Shift the contents of y) to Sr1.

At 8806, it is determined whether the status register ST3 is 0 point or D point, and if it is 0 point or D point, the contents of the X counter are stored in xREGI. At 5808, status registers STI, Sr1. The combination of states of Sr1 is determined and matches the combination shown in 5618. 5ell ~58
13 From 2, add xREGI to Xi.

Store the contents of the X counter in x2 and the contents of the X counter in yl. At 5814 and 5815, the X counter is incremented, and the process returns to 5801 and the same process is performed until it matches the lower side of the character circumscribing frame, but 5E108
If the combination is the latter two terms of 8618, then 58
11te stores xl plus -1 from the contents of xREGI.

At 8608, status registers 5TI-3T3 are
If the combination does not match the combination shown in 81G.

S809. At S810, the X counter is incremented until it matches the right side of the character circumscribing frame (5603).
Return to 5815 and repeat the above process. If the value of the X counter matches the lower side of the character circumscribing frame in 5815, xl.

The cutting point is determined at the value of X2*Yx. FIG. 9 is an example of a pattern when cut out according to this embodiment, and A
-A' indicates the division position of the pattern.

As explained above, according to this embodiment, even when the character pattern before and after 1 overlaps the character pattern, the character pattern can be changed without missing the character pattern or mixing in a part of the character pattern before and after 1. Can be cut out.

Furthermore, in this embodiment, the case where two characters overlap is shown, but even when three or more characters overlap, the same effect can be obtained by sequentially determining the cutting point based on the first two characters of the overlapping characters. can be obtained.

(Effects of the Invention) As described above, according to the present invention, by performing column scanning from the upper and lower sides of the circumscribing frame of a character pattern toward the opposite sides, the background portion is classified into areas according to the scanning direction, Based on the classification results, the lines within the circumscribed frame are scanned to detect the cutout area and the cutout position is determined, so that character cutout can be performed with high precision. Further, since this is realized by scanning the circumscribed frame of the pattern and detecting the change point, it can be implemented with a simple circuit configuration. moreover,
By using the present invention, it is possible to cut out even when adjacent characters overlap, so it is possible to reduce the interval between character entry frames and increase the number of readable characters per line. Therefore, it can handle many types of forms, has a large degree of freedom in form design, and has a high performance OC.
This has the effect of realizing R.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram showing a specific example of column scanning in this embodiment, FIG. 3 is a diagram showing the processing results of column scanning in FIG. 2, and FIG. Figure 4 is a diagram showing an example of the pattern in which the cutting position in Figure 2 has been determined;
The figure shows the configuration of the pattern memory in this embodiment.
Figure 6. 7 and 8 are flowcharts showing the processing flow of this embodiment, FIG. 9 is a diagram showing an example of a pattern when cropping is performed according to this embodiment, and FIG. It is a figure showing an example of a pattern. 200--One image signal, 201--Line buffer,
202-m-circumscribing frame creation circuit, 203-m-switching circuit,
204---Pattern memory, 205---x counter, 2013---7 counter, 20? -m-control circuit. 208--One stroke pattern area classification circuit, 209--One white point → black point change point detection circuit. 210---Pattern area change point detection circuit, 211-Completely exposed area detection circuit, 212---y RE G, 213---x R
E G■, 214--x RE G I. Patent Application Hitoki Electric Industry Co., Ltd. Patent Application Agent

Claims (1)

[Claims] In a character extraction method in which a quantized character pattern string obtained by photoelectrically converting a character string written on a form is separated and extracted character by character, the character pattern string is stored in a line buffer memory and means for creating a black point histogram in the column direction by operating the line buffer memory one column in the column direction of the character string, and detecting the width of the black point histogram in the column direction by sequentially performing the scanning for each column; means for detecting how many characters the width of the black dot histogram corresponds to by comparing the average character width of a specified target character with the width of the black dot histogram in the column direction; and based on the number of characters included in the black dot histogram. A black point histogram in the row direction is created by scanning the character pattern string in the region row by row in the row direction, and the character pattern string is determined from the black point histogram in the column and row directions. means for detecting a character circumscribing frame; storage means for retaining a character pattern string within the character circumscribing frame; a detecting means for reading the contents of the character pattern string from the accompanying storage means and detecting whether the contents are a character part or a background part; and a background part and a background part detected by scanning from the upper side obtained by the detecting means. a classification means for classifying a character pattern string within the character circumscribing frame into four types: a background portion detected by scanning from the lower side, a character portion, and a background portion other than these, and a character pattern string within the character circumscribing frame only; A change check in which the line scan is performed again to detect change points at which the classification by the classification means changes, and the change points are sequentially stored and sequentially compared with a predetermined combination of the change points to detect the matching change points. 1. A character cutting method comprising: output means; and means for determining a character cutting position based on the change point detected by the change point detection means.