JPH0795331B2 - Character cutting device - Google Patents

Description

TECHNICAL FIELD The present invention relates to a character cutting device, and more specifically, it reads characters written on a form, and a character pattern string based on the read characters is read for each character area. The present invention relates to a character cutting device that separates and extracts.

(Prior Art) In an optical character recognition device (hereinafter abbreviated as OCR), characters written on a form are scanned line by line, an optical signal is converted into an image signal by a photoelectric converter, and stored in a line buffer. . The line buffer is sequentially read out and the character pattern string is separated for each character area, and recognition is performed by the separated character pattern. Therefore, the character segmentation method for extracting one character area from the character pattern string is OCR. It greatly affects the performance.

In the OCR, one line is scanned from the upper end to the lower end of the line buffer in which the character string is stored, and the character pattern of the line buffer is read by sequentially moving the column in a direction perpendicular to this scanning. Further, a histogram is created by counting black dots (black dots in the character portion and white dots in the background portion) during scanning of one column, and the area of one character is determined by referring to the black dot histogram.

However, in the case of handwritten characters, the characters are entered obliquely, or the characters are out of the character entry frame, or because the writer has partially spelled out the characters. The letters to be overlapped,
There is a problem that a character pattern of two or more characters is cut out as one character.

In order to solve this problem, Japanese Patent Application No. 60-36574 proposed by the applicant of the present invention determines how many characters correspond to the width of the black dot histogram. The character circumscribing frame of the string is detected, and the character pattern string in the character circumscribing frame is held in the storage means. Next, scanning is performed from the upper and lower sides of the character circumscribing frame toward the opposite sides, the content of the character pattern string is read from the storage means, and it is detected whether the content is a character portion or a background portion.

In this way, when the background portion detected by the scanning from the upper side and the background portion detected by the scanning from the lower side, the character portion, and the character portion in the scanning are detected, the scanning of the column is terminated, and therefore, The character pattern sequence in the character circumscribing frame is classified into four types of background portions that have not been scanned. Next, horizontal scanning is performed to detect change points at which the classification changes, sequentially store the change points, and at the same time, hold the states (classification results) before and after the change points, and change the transition of the states to the transition of the change of a predetermined classification. The matching change point is detected by comparing with the combination of
Due to the change point, characters were cut out from a character pattern string in which a part of adjacent characters overlap.

FIGS. 9 (a) and 9 (b) are diagrams showing a specific example of the conventional character cutting method described above. In the figure, 100 and 101 are character patterns. 103 is the scanning direction from top to bottom, 1
04 indicates the scanning direction from the lower side to the upper side. As shown in FIG. 9 (a), scanning is performed from the upper and lower sides of the character circumscribing frame toward the opposite sides, and four types of character pattern rows are concerned (background part scanned from the upper side: in the figure. Indicated by "2". Background part scanned from the lower side: indicated by "4" in the figure. Background part not subjected to both scans: indicated by "0" in the figure Character line part: "1" in the figure Classification is “4” → “0” →
The area converted to "2" was detected, and the cutout position such as AA 'was detected from the change point.

(Problems to be Solved by the Invention) However, in the above-described conventional character cutout method, there is no combination of transitions of classification results in a character circumscribed frame that matches a predetermined combination. The line could not cut out one letter. In other words, when the characters fill out more than the entry frame or bounce off due to the customs or habits of the person who fills out the form, if a part of the character greatly enters into the adjacent character, or if it overlaps more complicatedly, As shown in FIG. 9 (b), the results of scanning from the upper and lower sides are scanned in the horizontal direction to detect change points at which the classification changes, hold the change points, and at the same time classify before and after the change points. There is a problem that it is not possible to hold the results and detect the combination of the classification results that matches the predetermined combination.

An object of the present invention is to solve these problems, and an object thereof is to provide a character cutting device having a simple structure and high accuracy.

(Means for Solving Problems) In order to solve the above problems, the present invention provides a character for each character from a quantized character pattern string obtained by photoelectrically converting a character string written on a form. In a character segmentation device for separating and extracting patterns, a line buffer memory for storing the character pattern sequence and a line-point black line histogram are created by scanning the line buffer memory column by column in the character pattern sequence. Then, a block detection unit that detects a block to be cut out based on the width of a black dot histogram in which black dot histograms larger than the first threshold continuously continue for the second threshold or more, and a width of the block obtained by the block detection unit. A determination means for comparing with the third threshold value to determine how many characters the width corresponds to, and a character pattern string of the block detected by the block detection means in the row direction. A circumscribing frame detecting means for detecting a character circumscribing frame on the basis of the row-direction black dot histogram created by scanning for each row and the column direction, and a character in the character circumscribing frame obtained by the circumscribing frame detecting means. When the storage means for storing the pattern sequence and the determination means determines that there are a plurality of characters, scanning is performed from the upper side and the lower side of the character circumscribing frame toward opposite sides, and the storage means stores the character pattern sequence. The content is read to detect whether the content is a character area or a background area, and by this detection processing, a background area detected by scanning from the upper side, a background area detected by scanning from the lower side, and a character area. , And when a character area is detected in the scan, the scan of the column is aborted,
A horizontal scanning of a classifying unit that classifies the character pattern string in the character circumscribing frame into each region of the background region that has not been scanned due to the suspension of the scan, and a classification result in the character circumscribing frame by the classifying unit. The change points that change the classification are detected and sequentially stored, the states before and after the change point are retained, and the transition of the state is compared with the combination of the transitions of the change of the predetermined classification, and the change points that match are changed. And a change point detecting means for detecting the change point detecting means and a change point detecting means does not detect a matching change point, the dividing means divides the character circumscribing frame in the horizontal direction, and each division area divided by the dividing means. Is used as a new character circumscribing frame, and the deciding means for deciding the character cut-out position on the basis of the changing point detected for each divided area by the classifying means and the changing point detecting means.

(Function) According to the present invention, since the character cutting device is configured as described above, the technical means functions as follows. The block detection means creates a black dot histogram in the column direction of the character pattern string in the line buffer memory to obtain the width of the black dot histogram in the column direction, and detects the block to which the character pattern string is cut out correspondingly. work. The judging means works to judge how many characters the detected block is. The circumscribing frame detection means creates a black dot histogram in the row direction of the character pattern string in the block, and detects the character circumscribing frame of the character pattern string in the block from this and the black dot histogram in the column direction. The storage means operates to store the character pattern in the character circumscribing frame. When the determining unit determines that the block has a plurality of characters, the classifying unit functions to scan from the upper and lower sides of the character circumscribing frame toward the opposite sides to classify into four types of regions. In response to this classification result, the change point detecting means horizontally scans to detect change points in which the area changes, sequentially stores them, holds the states before and after the change points, and changes the state of the change points of the change of a predetermined classification. It acts to compare transition combinations and detect matching transition points. Here, when no matching change point is detected, the dividing means works to divide the character circumscribing frame in the horizontal direction.
With respect to each of the divided areas, the changing points are detected by the classifying means and the changing point detecting means in the same manner as described above, and the determining means functions to determine the character cut-out position based on these changing points. Therefore, even when a part of a character deeply enters an adjacent character, the character can be cut out with high accuracy.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an 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, 202 is a black dot histogram creation circuit 220, a circumscribing frame detection circuit 221, and a character determination circuit 222. Reference numeral 203 is a data switching circuit, 204 is a pattern memory, and 205 and 206 are x-direction x-address counters (hereinafter referred to as x-counters) and y-direction y-address counters (hereinafter referred to as y-counters) that generate addresses for pattern memories. Reference numeral 207 is a control circuit. Reference numeral 208 is a pattern area classification circuit, and 209 is a character line detection circuit that detects a change point from a white point to a black point. Reference numeral 210 is a pattern area change point detection circuit, and 211 is a cutout area detection circuit. Further, a single arrow connecting each circuit indicates a data flow, and a double arrow indicates a control signal.

The operation of this embodiment will be described below with reference to the block diagram of FIG.

The character string on the form is converted into a binarized image signal 200 by a photoelectric converter and stored in the line buffer 201. The following processing is performed under the control of the control circuit 207. The control circuit 207 reads the image signal stored in the line buffer 201 from the head position of the line buffer 201 in column units, sequentially updates the columns, and ends when all the character pattern data for one line is read. Further, the control circuit 207 reads out the pattern data from the line buffer 201 on a column-by-column basis and, at the same time, activates the black dot histogram creation circuit 220. The black dot histogram creation circuit 220 creates a black dot histogram of the column by counting the number of black dots being read in one column, and stores the black dot histogram in the histogram memory 230 included in the black dot histogram creation circuit 220. The above processing is repeated, and the black dot histograms of all rows for one row are stored in the histogram memory 230.
The process ends when the data is stored in.

After the black dot histogram for one row is created, the histogram memory 230 in the black dot histogram creating circuit 220 is read from the beginning, and the block is detected by referring to the black dot histogram. The control circuit 207 sequentially reads out the black dot histogram from the histogram memory 230 in the black dot histogram creating circuit 220, compares the black dot histogram with the threshold value α (α: a constant, but in this embodiment, α = 1), and the histogram Is larger than the character block start point candidate, the storage addresses of the black dot histograms are sequentially incremented, and the columns in which the read black dot histogram is larger than the threshold value α are counted, and β (β: a constant, but in the present embodiment, When β = 2) consecutive columns, the starting point candidate is set as the starting point. Further, the column is further updated, and after the start point is detected, the row where the black spot histogram becomes smaller than the threshold value α for the first time is set as the end point, and the region indicated by the length from the start point to the end point is set as the block. Next, the control circuit 207 activates the character determination circuit 222 to set the detected block length to the thresholds γ ₁ , γ ₂ (γ ₁ ,
γ ₂ constant, provided that γ ₁ = 75 and γ ₂ = 12 in this embodiment.
5 and) and compare. Then, the length W of the block
Is smaller than the threshold value γ _{1, the} block is determined to be one character, when γ ₁ ≦ W ≦ γ ₂ , it is determined to be 2 characters, and when W> γ ₂ is determined to be 3 or more characters. Further, the control circuit 207 activates the circumscribing frame detection circuit 221 for the block after determining the block, and detects the circumscribing frame. Further, when the circumscribing frame of this block is detected, the character pattern in the circumscribing frame is transferred to the pattern memory 204. Here, if W> γ ₂ , that is, if it is determined that the block has three or more characters, the cutting process from the starting point to γ ₂ is performed to
Split the first and second characters. With the cut-out point of this result as the starting point, the cutting-out processing is performed from that starting point to γ _2, and further the cutting-out processing is performed such that the second character and the third character are divided, and processing is performed up to W. Here, the fifth described later
As shown in the figure, the upper left edge of the character circumscribing frame is the origin, the lower edge position is PB, and the right edge position is PR.

Next, the processing of a character that has been determined to be two or more by the above character determination will be described with reference to FIG.

The control circuit 207 sets the x counter 205 and the y counter 206 which give addresses of the pattern memory 204 to the left end position of the upper side of the character circumscribing frame and increments the y counter 106 to scan toward the lower side of the character circumscribing frame. To do. Then, the address of the pattern memory 204 is set to (x, y) for the X axis and the Y axis, and the values of the x counter and the y counter are used. P the contents of the pattern memory 204 at the position indicated by the address
Expressed as M (x, y). In this embodiment, the white point is PM (x,
y) = 0, black points are PM (x, y) = 1, white points detected during scanning from the upper side are PM (x, y) = 2, white points detected during scanning from the lower side are PM (X, y) = 4. Therefore,
The pattern memory 204 in this embodiment has a data width of 3 bits for one mesh. Pattern area classification circuit 20
In 8, the character pattern is read from the pattern memory 204 by setting an address at the upper left end of the character circumscribing frame. PM (x, y)
When = 0, (PM (x, y) .OR.2) is set as a new PM (x, y), and the corresponding address of the pattern memory 204 is written via the switching circuit 203.

The control circuit 207 detects the change point from the white point to the black point. When the character line detection circuit 209 detects the black point with PM (x, y) = 1, the scanning of the column is stopped and the x counter 205 is incremented by one. Then, the scanning of the next column is performed from the upper side of the character circumscribing frame. Further, when the scanning is performed from the upper side of the character circumscribing frame and the lower side is reached, the scanning of the column is stopped and the scanning of the next column is performed. The above scanning is sequentially repeated, and when the rightmost column of the character circumscribing frame is processed, the processing ends. When the scanning from the upper side is completed, the control circuit 207 sets the x counter and the y counter to the left end of the lower side of the character circumscribing frame, performs the scanning from the lower side to the upper side, and performs the same scanning as that 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). When the rightmost column is processed in the same manner as the scanning from the upper side, the process is completed. After the two types of scanning are completed and the patterns in the character circumscribing frame are completed, the control circuit 207 controls the x counter 205 and y counter.
The counter 206 is set to the left end of the upper side of the character circumscribing frame, and horizontal scanning is performed to detect the character cutout area.

Here, as a specific example of the above-mentioned column scanning, the above-described FIGS. 9A and 9B can be cited.

Next, the detection of the character cutout area will be described with reference to FIG.

First, when the pattern area change point detection circuit 210 is activated by the control circuit 207, it horizontally scans the circumscribed frame while reading character pattern data (classification result by vertical scanning) from the pattern memory 204. Further, the pattern area change point detection circuit 210
Processes the character pattern data from the pattern memory 204.

FIG. 2 is a block diagram showing the internal configuration of the pattern area change point detection circuit 210. In the figure, 300 is a y counter 2
06 output, 301 is x counter 205 output, 302 is character pattern data, 303 is x counter 205 register (REG
A) and 304 are registers (REG for holding character pattern data)
D), 305 are comparators for comparing character pattern data,
306 is a register (xREG I) for detecting the coordinate of the X axis one point before the point where the change point of the character pattern data is detected, and 307 holds the X coordinate by the signal 310 given from the cutout position judging circuit 211 described later. Register (xREG II) 308
Is a register (yREG I) for holding the Y coordinate when the change point is detected.

The operation of the pattern area change point detection circuit 210 will be described below with reference to FIG.

The character pattern data read by the horizontal scanning in the circumscribing frame is held in the character pattern data holding register (REGD) 304. Then, the stored character pattern data, that is, the character pattern data immediately before one point and the character pattern data currently read out are compared by the comparator 305, and if it is determined that the result is changed, the
The coordinate position in the X-axis direction before the point is held. That is, the output 301 of the x-counter is held in the register (REGA) 303 every time the coordinates are moved further. When the change point is detected by the comparator 305, the output 301 is held in the register (REGA) 303.
The coordinates before the point are held in the register (XREG I) 306.

Next, in the cutout area detection circuit 211 shown in FIG.
When the pattern area change point detection circuit 210 detects a change point in the pattern area, the PM (x-1, y) character pattern data (area) is held in the register.

The cut-out area detection circuit 211 has three status registers that hold the PM (x-1, y) character pattern data, and the status registers store the contents of the registers when a change point in the pattern area is detected. Are shifted to adjacent registers. Further, when the change point of the pattern area is detected and the shift of the status register is completed, it is detected whether the contents of the three kinds of status registers match a predetermined value.

FIG. 3 is a block diagram showing an internal configuration of the cutout area detection circuit 211. In the figure, 400 is a control signal from the control circuit 207, 401 is character pattern data from the pattern memory 204, and 402 is a change point detection signal detected by the pattern area change point detection circuit 210. Reference numerals 403 to 406 are transition registers (combination circuits) indicating predetermined combinations of transitions of the character pattern data from the control circuit 207. 407 to 409 are status registers (ST3 to ST1) which hold the contents of the character pattern data when the change point is detected by the pattern area change point detection circuit 210. 410 to 413 are status registers 407 to 409 and transition registers
It is a comparison circuit (comparator) that detects a match of 403 to 406. Reference numeral 414 is an OR circuit that logically ORs the outputs of the comparison circuits 410 to 413, and 415 is a flip-flop that holds that the transition registers 403 to 406 and the state register match.

Next, the operation of the cutout area detection circuit 211 will be described with reference to FIG. The character pattern data 401 is held in the status register (ST3) 407 every time the coordinate position is advanced. When the pattern area change point detection circuit 210 detects a change point in the pattern area, the output of the status register (ST3) 407 is
2) Held at 408. At the same time, the status register (ST
1) 409 holds the output of status register 408. As described above, combinations of changes in three consecutive character pattern data are held in the three status registers 407, 408, 409. When a change point is detected, three status registers 407-4
The combination of 09 and a predetermined number of transition registers 403 to 406 are compared by comparison circuits 410 to 413, respectively, and if there is a combination that matches, a flip-flop indicating that the cutout area has been detected.
Set 415.

Next, when the flip-flop 415 is set in FIG. 1, the decision signal from the cut-out area detection circuit 211 is (y REG) 308 and (x REGII) of the pattern area change point detection circuit 210.
Supplied to 307. At that time, the contents of the x counter 205 or the y counter 206 stored in each register are output as y ₁ and x ₂ from each register.

If the status registers 407-409 match the given combination,
The horizontal scanning of the row is stopped, the y counter 206 is incremented, and a new horizontal scanning of the next row is performed. The horizontal scanning above is x ₁ (x REG
The cutout position is determined based on I), x ₂ (x REG II), and y ₁ (y REGI).

The predetermined combination used here is "4" → "0" → "2", "2"
→ 4 types of "0" → "4", "4" → "2" and "2" → "4" are stored in the transition registers 403 to 406, respectively.

Status register when horizontal scanning is completed within the bounding box
If there is no match in 407 to 409 with the transition registers 403 to 406, the control circuit 207 obtains the horizontal bisector of the character circumscribing frame. The control circuit 207 detects the cutout position in each of the two circumscribing frames in the same manner as the slicing procedure, using the two regions divided by the horizontal bisector as new circumscribing frames. That is, first, for the circumscribing frame above the horizontal bisector, the inside and outside of the circumscribing frame are classified into four types by scanning from the upper and lower sides to the opposite sides, and the classification result is changed while performing horizontal scanning. The change point and the combination of classification transitions before and after the change point are detected, and the change point that matches the combination and the predetermined combination is detected and set as the cutout position of the circumscribed frame above the dividing line. The control circuit 207 then detects the cutout position in the same direction also in the area below the dividing line, and connects the cutout positions to obtain the cutout position of the entire character pattern.

When black dots are reached without being detected when scanning from the upper side to the lower side of the circumscribed frame that is divided here, the position of that scanning line is taken as the cut-out position, and when multiple scanning lines are detected, the character The scanning line closest to the center position of the circumscribing frame of the pattern in the x-axis direction is set as the cutout position. The position on the x-axis that gives the cut-out position is x _P.

As described above, the cutout position x ₁ x ₂ y ₁ or x _P in the upper and lower areas
When is detected, the two cutting lines that have been detected are connected to each other by horizontal dividing lines to form the entire cutting position.

The pattern transfer method will be described below using the pattern example in which the cutout position is determined as shown in FIG. Also, the fourth
The figure shows patterns and cut-out positions stored in the pattern memory 204 in the block diagram of FIG.
The coordinates are such that the horizontal axis is the X axis and the vertical axis is the Y axis, and the pattern memory 204 is located in the fourth quadrant. XM and Y
M indicates the size of the pattern memory 204, and in this embodiment, XM = YM = 128 meshes. PR and PB indicate the circumscribing frames of the patterns stored in the pattern memory 204, and the circumscribing frame is represented by four straight lines X = 0, X = PR, Y = 0, Y = PR. In FIG. 4, reference numerals 300 and 301 denote patterns, and straight lines Y = y ₁ , X = x ₁ , and X = x ₂ denote cutout positions. In the pattern memory in this embodiment, data representing one mesh has the structure shown in FIG. In the figure, (1)
When 1 is 1, it means that there was a white spot during column scanning from the lower side to the upper side, and when (1) is 0, it means that it was other than the white spot. Further, when (2) is 1, it means that there was a white spot during column scanning from the upper side to the lower side, and when (2) is 0, it means other than the white point. Furthermore, when (3) is 1, it means a black dot, and (3) is 0.
When, it means a white dot. Therefore, the pattern data to be transferred is only the data shown in (3). X
The mesh on the straight line represented by = 0 is Y from the point of Y = 0.
The pattern data is transferred to the point of Y = PB by incrementing the coordinates one by one. The X coordinate is incremented after the completion of the one-column transfer. Repeat the transfer for each column,
When the transfer of the column of = x ₁ is completed, the pattern data is masked and the fixed value 0 is transferred from the next column to the column of X = x ₂ from the Y coordinate 0 to Y _D. Up to the next column of X = x _3, the Y coordinate from 0 to Y _D and Y ₁ to PB are masked and a fixed value of 0 is transferred. The transfer of the pattern 300 is completed at the time when the transfer to the column of X = x ₃ is completed. The pattern 301 can also be transferred by the same method. Further, for data in which one character is included in the circumscribing frame, the pattern in the circumscribing frame can be transferred by a similar method.

Next, the flow of processing of this embodiment will be described 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 flow charts for classifying pattern regions and determining cut-out regions by scanning two times, respectively. First, the sixth
The flow chart of the entire figure will be described. In S400, the reading operation is started. In S401, one row of the pattern data stored in the line buffer is read out, a black dot histogram creating circuit 220 in FIG. 1 creates a black dot histogram and stores it in the histogram memory 230. In S402, the completion of creation of all the black dot histograms for one line is detected, and the process of S401 is repeated until all the lines are created. In S403, the processed characters are managed, and the following processes are repeated until the cutting out of all the characters in one line is completed. In S404, the black dot histogram is read from the histogram memory, and the start point and end point of the black dot histogram are detected and set as blocks. Further, the length of the block is compared with the thresholds γ ₁ and γ ₂ and the number of characters in the block is held. In S405, the circumscribing frame detection circuit 221 of FIG. 1 detects the circumscribing frame of the block,
The pattern data in the circumscribed frame is transferred to the pattern memory 204. In S406, the pattern data of the pattern memory 204 is transferred to the output stage if there is one character according to the judgment result of the length of the held block, and the process proceeds to the next character. Two
If it is more than the character, the following processing is performed. In S407, column scanning is performed from the upper side and the lower side of the circumscribing frame to the opposite side, the pattern areas are classified, and the result is recorded in the pattern memory 204.
To store. In S408, horizontal scanning is performed within the circumscribing frame, the classification result is read from the pattern memory 204, and the cutout area is detected to determine the cutout position.

In S411, it is detected whether or not the cutout position can be determined. If detected, in S409, the pattern in the pattern memory 204 is transferred according to the cutout position, and the pattern memory 204
When all the patterns in the above are transferred, the next character is processed. If the cutout position cannot be detected, the following processing is performed. In S412, a horizontal bisector is detected for the circumscribing frame of the block, and in S413, a new upper circumscribing frame divided by the horizontal bisector is obtained. S414, S41
In step 5, the same processing as in S407 and S408 is performed on the specified circumscribed frame, and in step S416 it is detected whether the processing of both the upper and lower circumscribed frames has ended. If the lower side has not ended, the lower circumscribed frame in S418. Is detected, and the lower cutout region is detected in S414 and S415. When the cutout position is determined by the upper and lower circumscribed frames
In S417, the block cutout positions are combined from the upper and lower cutout positions, and the pattern in the pattern memory 204 is transferred in S409. When all the patterns in the pattern memory 204 have been transferred, the next character is processed.

Next, the processing of S407 and S408 in FIG.
A detailed flowchart is shown in FIG. 8 and FIG. 8, and its operation will be described in order.

FIG. 7 shows the flow of classification of character pattern areas and detection of change points from white points to black points. When character pattern data is input in S500, initialization is performed in S501 to S503, and the x, y coordinates of the pattern memory 204 are set at the upper left end of the character circumscribing frame, and the value U / D indicating the scanning direction is set. Since the scanning is performed from the upper side toward the lower side, it is set to 2. In S504, the contents of the pattern memory are checked, and if PM (x, y) = 1 (black dot), the process is executed in S514.
If PM (x, y ≠ 1 (white), then the contents of the pattern memory 204 are PM (x, y) = (PM (x, y) .OR.U / in S505.
D). In S506, the value of the y counter is incremented or decremented depending on the scanning direction. In S507, the end of one example is detected, when the end of one example is detected, the processing of S508 and subsequent steps is performed, and when the end of one example is detected, S5 is performed.
08 The following processes are performed, and if one example is not completed, the process returns to S504 and the above processes are repeated. When the end of one row is detected in S507, 1 is set to E which is the end of one row flag in S508. In S509, if the column is the first column due to the flag C indicating that it is an example in which the end of one column is detected for the first time, in S510, the X counter of the column is
The value of 205 is stored in x _S, and when the previous column is also the column in which the end of one column is detected, the process proceeds to S511. In S511, the flag C is set to 1 because the column has detected the end of one column, and the flow advances to S512 to process the next column. In S514, the one-column end detection flag C is determined, and if it is 0, the process proceeds to S516, and if the flag C is 1, 1/2 (x _S + x−1) is registered in x _P as a cutout point. All the detected x _P are registered and used as the cutout position when the cutout positions of S417 are combined in FIG. S516
In, the flag C is reset, and the process proceeds to S512 to perform the process of the next column.

When it is detected in S517 that the x counter 205 is at the right end of the circumscribing frame, if the one-row end flag E has been detected in S518, the next process is terminated and in S521, the pattern region classification process by the character line detection is performed. finish. If flag E has not detected the end of one row, U / D in S519 = U / D in S520 if 2
Is set to 4, the x counter 205 and the y counter 206 are initialized, and the process returns to S504 for processing. When U / D = 4, the process proceeds to S521 and ends the processing.

FIG. 8 also shows the flow for detecting change points of the pattern area and determining the cutout area. The horizontal scan in the character circumscribing frame of the pattern memory 204 is performed from the left end of the upper side to determine the cutout area. In S600 and S601, the y counter 206 and x counter 205 are initialized, respectively. In S602, the state registers (ST1 to ST3) 409 to 407 for holding the changes in the area during row scanning are initialized. The state of the current position is ST3, the region before the current region being scanned is ST2, and the region before ST2 is ST1. In S603, the contents PM (x, y) of the pattern memory 204 are compared with ST3, and if they match, the process proceeds to S605. If they do not match, the coordinate is the change point, so the contents of ST2 and ST3 are changed in S604. Shift to ST1 and ST2 respectively. In S605, the contents of PM (x, y) are shifted to ST3. In S606, it is determined whether the status register ST3 is "2" or "4". When the status register ST3 is "2" or "4", the content of the x counter 205 is stored in (x REGI) 308. In S608, the state combination of the status registers ST1, ST2, ST3 is determined, and S61
If they match the combinations shown in 6, then use S611 to S613 to determine x
₁ stores (x REGI) 306, x ₂ stores the contents of the x counter 205, and y ₁ stores the contents of the Y counter 206. In steps S614 and S615, the y counter 206 is incremented, the process returns to step S601 and the same processing is performed until it matches the lower side of the character circumscribing frame. However, if the combination in S608 is the second half of S616,
In S611, stores the plus -1 from the contents of x REGI to x _1.

In S608, if the status registers ST1 to ST3 do not match the combination shown in S616, the x counter 205 is incremented in S609 and S610, and the process returns to S603 until it matches the right side of the character circumscribing frame, and the above processes are repeated. Y counter 20 with S615
If the value of 6 matches the bottom of the character bounding box, x ₁ , x ₂ , y ₁
The value of is the cut-out point.

As described above, according to the present embodiment, even when the character patterns before and after overlap with the character pattern, the character pattern is omitted without missing the character pattern or mixing a part of the character patterns before and after the character pattern. It can be cut out.

Further, in this embodiment, the case where two characters overlap each other is shown, but even when three characters or more overlap each other, the cut-out points are sequentially determined based on two characters from the beginning of the overlapped characters. By doing so, a similar effect can be obtained.

(Effects of the Invention) As described above, according to the present invention, the background portion is classified into the regions by the scanning direction by performing the column scanning from the upper and lower sides of the circumscribing frame of the character pattern toward the opposite sides, According to the classification result, horizontal scanning in the circumscribing frame is performed to detect the cutout area and the cutout position is determined, so that highly accurate character cutout can be performed. Further, since the change point is detected by scanning the inside of the circumscribing frame of the pattern, it can be implemented with a simple circuit configuration. Furthermore, by using the present invention, it is possible to cut out even when adjacent characters overlap each other, so that the interval between the character entry frames can be reduced and the number of readable characters per line can be increased. Therefore, it can handle many types of forms,
There is a great degree of freedom in designing the form, and therefore OCR with good performance can be realized.

[Brief description of drawings]

1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a block diagram showing the internal structure of the pattern area change point detection circuit of FIG. 1, and FIG. 3 is a cutout area detection circuit of FIG. FIG. 4 is a block diagram showing an internal structure, FIG. 4 is a pattern diagram in which the cutout position of the embodiment of FIG. 1 is determined, FIG. 5 is a block diagram of data of one mesh of the pattern memory, and FIGS. First
FIG. 9 is a flowchart showing a character cutting process of the embodiment shown in FIG. 200 ... Image signal, 201 ... Line buffer, 202 ... Outer frame creation circuit, 203 ... Switching circuit, 204 ... Pattern memory, 205 ... x
Address counter (x counter), 206 ... y Address counter (y counter), 207 ... Control circuit, 208 ... Pattern area classification circuit, 209 ... Character line detection circuit, 210 ... Pattern area conversion point detection circuit, 211 ... Cutout area detection Circuit, 300 ... y counter output, 301 ... x counter output, 302 ... character pattern data, 303 ... register (REGA), 304 ... register (REG
D), 305 ... Comparator, 306 ... Register (xREG I), 3
07 ... Register (xREG II), 308 ... Register (yREG I),
400 ... Control signal, 401 ... Character pattern data, 402 ... Change point detection signal, 403-406 ... Transition register, 407-409 ... Status register, 410-413 ... Comparator, 414 ... Logical sum circuit, 4
15 ... Flip-flop (F / F).

Claims (1)

[Claims] 1. A character segmentation device for separating and extracting a character pattern for each character from a quantized character pattern sequence obtained by photoelectrically converting a character string entered on a form, wherein the character pattern sequence is A line buffer memory for storing the line buffer memory, and scanning the line buffer memory for each column in the column direction of the character pattern column to create a black dot histogram in the column direction.
Block detection means for detecting a block to be cut out based on the width of a black dot histogram in which a black dot histogram larger than the threshold value continues for a second threshold value or more, and a block width and a third threshold value obtained by the block detection means. And a determination means for determining how many characters the width corresponds to, and a black dot in the row direction created by scanning the character pattern string of the block detected by the block detection means in each row in the row direction. A circumscribing frame detecting means for detecting a character circumscribing frame based on the histogram and the black dot histogram in the column direction, a storage means for storing a character pattern sequence in the character circumscribing frame obtained by the circumscribing frame detecting means, and the judging means. Is determined to be a plurality of characters, the contents of the character pattern string are read from the storage means by scanning from the upper side and the lower side of the character circumscribing frame toward the opposite sides. Then, it is detected whether the content is a character area or a background area, and by this detection processing, a background area detected by scanning from the upper side, a background area detected by scanning from the lower side, a character area, and When a character region is detected by scanning, the scanning of the column is stopped, and the character pattern sequence in the character circumscribed frame is classified into each region of the background region that has not been scanned due to the termination of the scan. And the classification result in the character circumscribed frame by the classification means,
The horizontal scanning is performed to detect change points at which the classification changes, and the changes are stored sequentially, and the states before and after the change points are held, and the transitions of the states are compared with a combination of transitions of changes of a predetermined classification to be matched. Change point detecting means for detecting a change point, when no change point is detected by the change point detecting means, a dividing means for dividing the character circumscribing frame in a horizontal direction, and A character cutout device having a divided area as a new character circumscribing frame, and a determination means for determining a character cutout position based on a change point detected for each divided area by the classification means and the change point detection means. .