JP2000339408A - Character segment device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To more precisely judge the number of characters in a character block where plural characters are contained by using a relation between the character and a frame line and a position relation with the characters before and behind in a character segment device. SOLUTION: A frame line detection part 2 extracts a frame line of a whole input picture and a character block separation part 3 separates it into respective character blocks. A character number judgment part 4 judges the number of characters contained in the character block by using not only peripheral information from a circumscribing rectangle to the character but also the frame line detected by the frame line detection part 2 and the position relation with the character blocks separated by the character block separation part 3. A segment part 6 segments the character in a unit of one character through a noise removal part 5.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

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

[0002]

2. Description of the Related Art As a method for determining the number of characters in a character block in a conventional character cutout device, there is an "optical character reading device" disclosed in JP-A-6-33089 (hereinafter referred to as a first conventional character cutout device). In this “optical character reading device”, the number of characters in a character block is determined as follows. FIG. 27 shows the main configuration. Image input unit 2
Numeral 01 optically scans a form in which characters are written and takes in character image data into a memory. The projection unit 202 creates vertical projection data from the image data captured in the memory. The judging unit 203 judges from the projection data created by the projecting unit 202 whether it is a continuous character or a normal character (FIG. 28 (a)). The deletion unit 204 determines this judgment unit 20
When it is determined in step 3 that the character is a continuation character, an unnecessary portion as a character is deleted (FIG. 28B). The character cutout unit 205 cuts out one character at a time when the determination unit 203 determines that it is a normal single character, and a character from which an unnecessary part has been deleted by the deletion unit 204, and performs character recognition by the character recognition unit 206.

[0003] As a noise removing method in a conventional character extracting device, there is a "mail sorting device" disclosed in JP-A-8-16717 (hereinafter referred to as a second conventional character extracting device). In the "mail sorting apparatus", noise is removed as follows. FIG. 29 shows the main configuration. The position and area of each connected block of the graphic pattern normalized by the normalization unit 211 and blocked by the blocking unit 212 are calculated by the position area calculation unit 213, and the area ratio for each block is calculated from the calculated area. The area ratio calculation unit 214 calculates the area ratio as a noise determination criterion corresponding to the position of each block from the noise determination table 215, and compares the read area ratio with the noise determination unit 216 to determine whether the noise is present. The determined noise is removed from the graphic pattern normalized by the noise removing unit 217.

[0004]

However, in the first conventional character segmentation device, when the preceding and succeeding characters are connected at a plurality of positions as shown in FIG. The number of characters is determined to be smaller than that of the handwritten “4” with the top opened as shown in FIG.
In such a case, since the projection becomes minimal in the middle of the character, the number of characters is determined to be larger than the actual number.
There were challenges. Further, the second conventional character segmentation apparatus has a second problem that large noise as shown in FIG. 32 cannot be removed.

[0005]

According to the present invention, in order to solve the above first problem, a first number of temporary characters is obtained based on a positional relationship between a vertical frame and a character, and the first number of temporary characters is calculated. By calculating the number of characters in the final character block based on the base, it is possible to project and characterize two or more characters connected at multiple points included in multiple frames or handwritten “4” with an open top. Even if the number of characters is incorrect due to the determination of the number of characters of only the height information, the number of characters can be corrected in consideration of the balance between the entry frame and the characters, and more accurate number of characters can be determined.

In order to solve the second problem, the present invention is based on the premise that a handwritten character string is not included in a character string (that is, a handwritten determination character is noise). Using the result of handwriting determination, when a character block existing around the target character block is a type determination, the target character block determined by handwriting is determined to be noise and removed to remove only small noise. In addition to the removal, it is also possible to remove large noise contained in the character string.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The invention according to claim 1 of the present invention is directed to a case where a slip in which a character group is written in a horizontally continuous non-separable frame is input as a binary image. The binarization unit to be converted, and detects the coordinates of all vertical frame lines included in the image binarized by the binarization unit, adds numbers in order, and sets the coordinates and numbers of each frame line. , A character block separating unit that separates the image binarized by the binarizing unit into character blocks, and a character block separated by the character block separating unit. A character number determination unit that determines the number of characters, a noise removal unit that removes noise from the character lump group separated by the character lump separation unit, and a character lump group that has not been removed by the noise removal unit. A cutout unit that cuts out characters by the number of characters determined by the character number determination unit, The number-of-characters determination unit calculates the number and the number of the frame line included in the character block from the coordinates of the frame line detected by the frame line detection unit and the coordinates of the character block separated by the character block separation unit. The frame count calculation unit, and a frame line closest to the left end of the character chunk separated by the character chunk separation unit and the character in the frame group corresponding to the frame number calculated by the frame count calculation unit. A left frame line distance calculation unit for calculating a distance to the left end of the character block separated by the block separation unit; and a frame line group corresponding to the frame line number calculated by the frame content number calculation unit. A right frame line distance calculating unit that calculates the distance between the frame line closest to the right end of the character block separated by the unit and the right end of the character block separated by the character block separating unit; , The distance between the frame line detected by the left frame line detection unit and the left end of the character block, and the number of frame lines detected by the right frame line detection unit. A frame line character number judging unit for judging the first provisional number of characters of the character chunk from the distance between the frame line and the right end of the character chunk, and a vertical line from the top of the rectangle circumscribing the character chunk separated by the character chunk separating unit. An upper end distance calculation unit that calculates the distance to the character chunk for each, and a lower end distance calculation unit that calculates the distance to the character chunk for each vertical line from the lower end of the rectangle circumscribing the character chunk separated by the character chunk separation unit, From the distance from the upper end of the rectangle to the character block for each vertical line calculated by the upper end distance calculator and the distance from the lower end of the rectangle to the character block for each vertical line calculated by the lower end distance calculator, a second character block is obtained. The number of characters in a character block using the vertical character number determining unit for determining the number of temporary characters, the first temporary character number determined by the frame line character number determining unit, and the second temporary character number determined by the vertical distance character number determining unit First comparison character count And a character cutout device that includes the number of characters determined by the first comparison character number determination unit as an output of the character number determination unit. Basically, each character and each entry frame are in a one-to-one correspondence. Under the premise that they are supported, the frame line character number determination unit obtains the first provisional character number based on the positional relationship between the character block and the frame line, and the upper and lower distance character number determination unit determines The second provisional character number is obtained from the background distance from the lower end, and the first comparison character number determination unit determines the first provisional character number using the second provisional character number.
By adjusting the number of characters obtained by correcting the provisional number of characters to the final number of characters, the projection and character heights that occur when the preceding and following characters are connected or in the case of handwritten "4" with an open top By correcting the character number error due to the character number determination method using only information in consideration of the balance between the existing positions of the entry frame and the character, it is possible to more accurately determine the character number.

In the invention according to a second aspect of the present invention, the character number judging section uses the positional relationship between the target character block and the preceding and following character blocks for each character block separated by the character block separating section. The number of characters before and after the mass, which corrects the first provisional character number obtained by the frame line character number determination unit to obtain the third provisional character number, and replaces the first comparative character number determination unit with the vertical character number A second comparison character number determination unit that determines the number of characters in a character block by using the second provisional character number determined by the determination unit and the third provisional character number determined by the front and rear lump character number determination unit; 2. The character cutout device according to claim 1, wherein the number of characters determined by the comparison character number determination unit is output from the character number determination unit, based on the assumption that there is originally only one character in one frame. The frame where the left edge of the character block exists in the front and rear block character number determination unit Other (and previous character lump) character lumps, or else in a frame character lumps rightmost exists character lumps (
Is determined, it is determined whether the target character block or the character block immediately before or immediately after is the character to be included in the frame, and the second character block is determined based on the result. Correcting the number of characters in the character chunk by the comparison character number judging unit makes it possible to judge the number of characters in consideration of the balance of the number of characters with the surrounding character chunks. As a result, the accuracy of the character number judgment is improved.

According to a third aspect of the present invention, there is provided a binarizing section for binarizing an input image, and a character chunk separation for separating the image binarized by the binarizing section into character chunks. Part, a character number determining unit that determines the number of characters included in each character block separated by the character block separating unit, and a character block determined to be noise in the character block group separated by the character block separating unit. A noise removing unit that removes, and a cutout unit that cuts out characters for each of the character lump groups remaining without being removed by the noise removing unit by the number of characters determined by the character number determining unit, wherein the noise A removing unit configured to determine a small character block as noise in each of the character blocks separated by the character block separating unit, and to remove each of the character blocks separated by the character block separating unit; A character type determining unit for determining the type and handwriting of the character, and the character type A character type noise removing unit that removes the target character block when the target character block is handwritten in the fixed unit and the character block before and after the target character block is printed in the character type determination unit. It is a character cutout device, and based on the assumption that handwritten characters are not mixed in the printed character string, the character type determination unit performs type and handwriting determination for each character block, so that the target character block can be determined by handwriting. If the character chunk before and after the target is type determination, the target character chunk is determined to be noise, so that even large noise that cannot be removed by a normal small noise removal method is included in the type string. It has the effect that it can be removed.

According to a fourth aspect of the present invention, in the case where a slip in which a plurality of characters are continuously written in a horizontal direction is imaged and input, the character type determination unit determines whether the target character block is A height determination unit that determines the target character block as a print when the height is similar to the preceding and following character blocks, and the vertical coordinate of the upper end of the target character block approximates the vertical coordinate of the upper end of the preceding and following character blocks. Or, when the vertical coordinates of the lower end of the target character block are similar to the vertical coordinates of the lower ends of the preceding and following character blocks, a coordinate determining unit that determines the target character block as a printed type, and the height determining unit or the coordinate determining unit. 4. The character segmentation device according to claim 3, further comprising: a handwriting determination unit that determines a character block that is not determined to be a handwritten by the unit as handwriting. Even if the type is not clear, By it can be determined that the print in only the vertical coordinates of the print determination character lumps and upper or lower ends of the front and rear is approximated by a determination unit, an effect that than just the height determination, can determine the number of printed characters print with.

According to a fifth aspect of the present invention, in the case where a slip in which a plurality of characters are sequentially written in a horizontal direction and a character entry area is separated in units of fields is imaged and input, the characters are input. The type determination unit determines, for each character chunk in the same field, a character chunk whose height or vertical coordinate is similar to other character chunks in the same field as a print type, and determines all others as handwriting. 5. The character cutout device according to claim 4, further comprising a field determination unit, wherein basically all characters in the same field are composed of regularly arranged typefaces of the same font or handwritten. By determining the type and handwriting using the height / coordinate information of the character chunk, the handwriting / type determination can be performed using as much information as possible.
The accuracy of handwritten / printed character determination is increased, and the character block that has been determined by handwriting, that is, more noise, contained in the typed character block can be found and removed.

According to a sixth aspect of the present invention, in the case where a slip in which a character group is entered in a non-separable frame is input as an image, the input image is binarized and separated for each character block. Then, the coordinates of all vertical frame lines included in the binary image are detected and numbered, the coordinates and numbers of each frame line are stored, and the coordinates of the detected frame lines are stored for each character block. And, from the coordinates of the separated character block, the number and number of the frame lines included in the character block are calculated, and from the calculated frame line numbers, the number of the frame line closest to the left end of the character block and the character The number of the frame line closest to the right end of the chunk is determined, the number of the frame lines included in the character block, the left end coordinates of the character block, and the frame line closest to the left end of the character block among the frame lines included in the character block. Of the character block and the distance between the coordinates of the right end of the character block and the frame line closest to the right end of the character block among the frame lines included in the character block. The first number of provisional characters is determined, and for each rectangle circumscribing each character block, the distance from the upper end of the rectangle to the corresponding character block and the distance from the lower end of the rectangle to the corresponding character block are determined for each vertical line. The second provisional character number of the character block is determined from the magnitude and change of the distance of the character block, the actual number of characters of the character block is determined from the first provisional character number and the second provisional character number, and the character determined as noise from the character block group Remove lumps,
This is a character extraction method for executing character extraction for each character chunk group remaining without being removed by the determined number of characters. Basically, each character and each entry frame are 1
Under the premise that the character block corresponds to one, the first temporary number of characters is obtained based on the positional relationship between the character block and the frame line, and the second temporary character number is determined based on the background distance from the upper and lower ends of the circumscribed rectangle of the character block. By calculating the number of characters and correcting the first number of temporary characters with the second number of temporary characters to obtain the final number of characters,
In the case where the preceding and following characters are connected or the upper part is open, such as handwritten "4", the character number error by the character number determination method based on only the projection and character height information is calculated based on the position of the entry frame and the character. By performing the correction in consideration of the balance, there is an effect that the number of characters can be more accurately determined.

According to a seventh aspect of the present invention, when a voucher in which a character group is entered in a non-separable frame is input as an image, the input image is binarized and separated for each character block. Then, the coordinates of all the vertical frame lines included in the input image are detected and numbered, the coordinates and numbers of each frame line are stored, and for each character block, the coordinates of the detected frame lines and From the coordinates of the separated character chunk, calculate the number and number of frame lines included in the character chunk, and from the calculated frame line numbers, select the number of the frame line closest to the left end of the character chunk and the character chunk The number of the frame line closest to the right end is obtained, and the number of the frame lines included in the character block, the coordinates of the left end of the character block and the frame line closest to the left end of the character block among the frame lines included in the character block are determined. From the distance and the distance between the right end coordinate of the character block and the frame line closest to the right end of the character block among the frame lines included in the character block, the character block The first number of provisional characters is determined, and for each rectangle circumscribing each character block, the distance from the upper end of the rectangle to the corresponding character block and the distance from the lower end of the rectangle to the corresponding character block are determined for each vertical line. The second provisional character number of the character block is determined from the magnitude and change of the distance of the distance, the third provisional character number is determined from the positional relationship between the target character block and the preceding and following character blocks, and the first provisional character number, the second provisional character number, The actual number of characters of the character block was determined from the number of provisional characters and the third provisional character number, and the character blocks determined as noise were removed from the character block group. For each character block group remaining without being removed, the determination was made. This is a character cutout method that cuts out characters by the number of characters. Based on the premise that there is originally only one character in one frame, another character block (just before Or the right end of the character block If another character block (the following character block) is included in the existing frame, determine whether the target character block or the character block immediately before or after is the character to be included in the frame. Then, by correcting the number of characters in the character block based on the result, the number of characters can be determined in consideration of the balance of the number of characters with the surrounding character blocks, and as a result, the accuracy of the character number determination is improved.

According to an eighth aspect of the present invention, an input image is binarized and separated into individual character blocks, and the coordinates of all vertical frame lines included in the binary image are detected and numbered. After storing the coordinates and the number of each frame line, and determining the number of characters for each character block, the type and handwriting of each character block are determined, and the determination result of the target character block is handwritten, And, when the judgment result of the character block around the target character block is a print type,
The target character block is determined to be noise and removed, and the small character block in the character block group is determined to be noise and removed. For each of the character blocks remaining without being removed, only the determined number of characters This is a character segmentation method that executes character segmentation.Assuming that handwritten characters are not mixed in a typescript string, the type and handwriting judgment is performed for each character chunk to determine whether the target character chunk is handwritten. In the case where the character chunk before and after the target is a type determination, the target character chunk is determined to be noise, so that even large noise that cannot be removed by a normal small noise removal method is included in the type string. It has the effect of being able to be removed.

According to a ninth aspect of the present invention, there is provided a recording medium storing a program that realizes the character extracting method according to any one of the sixth to eighth aspects by software. Can be easily introduced into other systems, and by combining it with a system that has the function of a recognition unit or a recording medium, it is possible to simply read an input pattern and read characters contained in that pattern. Has the effect of becoming possible,
In addition, there is an effect that the burden of confirmation and correction on the user side is reduced as a result of the improvement in recognition performance.

(Embodiment 1) Hereinafter, a first embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram of a character segmenting apparatus according to the first embodiment of the present invention, and FIG. 24 shows an operation procedure thereof. FIG.
In FIG. 24, reference numeral 1 denotes a binarization unit, which performs binarization when an input image is input in multiple values (101 in FIG. 24). Reference numeral 2 denotes a frame detection unit which stores coordinates and numbers of vertical frame lines included in the image binarized by the binarization unit 2 (1 in FIG. 24).
02). Numeral 3 denotes a character chunk separation unit, which separates each character chunk from the image binarized by the binarization unit 2 from vertical and horizontal projections (103 in FIG. 24). Reference numeral 4 denotes a character number determination unit that determines the number of characters included in each character block separated by the character block separation unit 3. 5 is a noise removing unit,
Noise determination is performed on each character chunk separated by the character chunk separation unit 3, and the character chunk determined to be noise is removed (112 in FIG. 24). Reference numeral 6 denotes a cutout unit, which cuts out a character image from each character block remaining without being removed by the noise removal unit 5 by the number of characters determined by the character count determination unit 4 (113 in FIG. 24). .

FIG. 2 shows a block diagram of the number-of-characters judging section 4 in the present embodiment. Reference numeral 11 denotes a frame content calculating unit which uses the coordinates and numbers of the frame lines detected by the frame line detecting unit 1 to include each character block separated by the character block separating unit 3 in the character block. The number and the number of the frame line to be calculated are calculated (104 in FIG. 24). Reference numeral 12 denotes a left frame line distance calculation unit, which includes the left end of the character block separated by the character block separation unit 3 and the frame number calculation unit 1
In the frame line group corresponding to the frame line number calculated in step 1, the distance to the frame line closest to the left end of the character block is obtained (105 in FIG. 24). Reference numeral 13 denotes a right frame line distance calculation unit, which is included in the right edge of the character block separated by the character block separation unit 3 and the frame line group corresponding to the frame line number calculated by the frame number calculation unit 11. The distance to the frame line closest to the right end of the block is obtained (10 in FIG. 24).
6). Reference numeral 14 denotes a frame line character number determination unit, and the frame number calculation unit 1
If the distance obtained by the left end frame line distance calculation unit 12 is short, 1 is subtracted from the number obtained by adding 1 to the number of frame lines included in the character block obtained in 1, and the right end frame line distance calculation unit 13 If the distance obtained in is short, the first provisional character number is obtained by further reducing 1 (107 in FIG. 24). Reference numeral 15 denotes an upper end distance calculation unit, which calculates the distance from the upper end of the rectangle circumscribing the character chunk separated by the character chunk separation unit 3 to the character chunk for each vertical line (108 in FIG. 24). Numeral 16 denotes a lower end distance calculating unit, which obtains a distance from the lower end of the rectangle circumscribing the character block separated by the character block separating unit 3 to the character block for each vertical line (109 in FIG. 24). Reference numeral 17 denotes an upper / lower distance character number determination unit which calculates the distance from the upper end of the rectangle for each vertical line calculated by the upper end distance calculation unit 15 to the character block and the character from the lower end of the rectangle for each vertical line calculated by the lower end distance calculation unit 16 Using the distance to the chunk, a second number of provisional characters is obtained from the magnitude and change of the distance (110 in FIG. 24).
Reference numeral 18 denotes a first comparison character number determination unit which calculates the number of characters in a character block from the first provisional character number obtained by the frame line character number determination unit 14 and the second provisional character number obtained by the upper and lower end distance character number determination unit 17. It is determined (111 in FIG. 24).

The operation of the character segmenting apparatus configured as described above will be described below with reference to FIGS. As an example of the input image, a slip in which characters are entered in a horizontally non-dropout color frame as shown in FIG. 3 is used. In the following description, the horizontal direction is x and the vertical direction is y.

First, after the vertical frame lines are detected by the frame line detecting section 1, a number is added to each vertical frame line in order from the left and the x coordinate is also stored simultaneously (FIG. 4). Next, in the binarizing unit 2, after binarizing the input image, the character block separating unit 3
, The entire projections in the x and y directions are obtained, and the area where the projections exist is defined as a character block (FIG. 5). As a method of determining a character block, a method of labeling black pixels and forming each label as a character block may be used instead of projection.

Next, the number-of-characters determination unit 4 determines the number of characters for each character block. Hereinafter, the detailed operation of the number-of-characters determination unit 4 will be described with reference to the configuration of FIG. First, in the frame content calculating unit 11, for each character chunk separated by the character chunk separating unit 3, the range of the x direction coordinate where the character chunk exists, and each vertical frame line detected by the frame line detecting unit 1. And if the x-coordinate of the vertical frame is included in the range of the x-coordinate where the character block exists, the number and x-coordinate of the vertical frame and the total number of included vertical frames are stored. (FIG. 6). Next, in the left frame line distance calculation unit 12, the smallest number (leftmost frame line) among the frame line numbers calculated by the frame number calculation unit 11 is selected, and the x coordinate of the selected frame line is selected. ,
The distance from the x-coordinate of the left end of the character chunk separated by the character chunk separation unit 3 is obtained (FIG. 7). Further, the right frame line distance calculation unit 13 selects the largest number (rightmost frame line) among the frame line numbers calculated by the frame number calculation unit 11, and calculates the x coordinate of the selected frame line, The distance from the x-coordinate of the right end of the character chunk separated by the character chunk separation unit 3 is obtained (FIG. 8). Next, in the frame line character number determination unit 14, the left end frame line distance calculation unit 12 calculates the number based on the number obtained by adding 1 to the number of frame lines included in the character chunk obtained by the frame number calculation unit 11. When the distance is shorter than a predetermined threshold, the first temporary character number is obtained by subtracting 1 when the distance obtained by the right end frame line distance calculation unit 13 is short (FIG. 9).

Next, the upper end distance calculator 15 calculates a vertical distance (upper distance) from the upper end of the rectangle circumscribing the character chunk separated by the character chunk separator 3 to the character chunk for each x-coordinate (FIG. 2). 10). Further, the lower end distance calculation unit 16 obtains a vertical distance (lower end distance) for each x coordinate from the lower end of the rectangle circumscribing the character lump separated by the character lump separation unit 3 to the character lump (FIG. 11). The upper and lower distance character number determination unit 17 uses the upper end distance for each x coordinate obtained by the upper end distance calculation unit 15 and the lower end distance for each x coordinate obtained by the lower end distance calculation unit 16 to determine the point at which each distance is large. A point having a large change in the distance at the adjacent x coordinate, or a point having a small character height obtained by subtracting the upper end distance and the lower end distance from the height of the circumscribed rectangle is set as a division candidate point between characters. A value obtained by adding 1 to the number of points is set as a second provisional character number (FIG. 12).
However, when the distance between the division candidate points is short, one of the division candidate points is deleted, and the second provisional character number is reduced accordingly. As a criterion of deletion, priority is given to each candidate division point according to the calculation method of the division candidate points (bottom distance, top distance, character height), the size of the distance, the character height, and the distance from the frame line May be added, and a division candidate point having a lower priority may be deleted when a division candidate point is deleted.

The first comparison character number determination unit 18 calculates the number of characters of the character block from the first provisional character number determined by the frame line character number determination unit 14 and the second provisional character number obtained by the vertical distance character number determination unit 17. judge. The rules for determining the number of characters include, for example, the following. (1) When the first provisional character number = the second provisional character number • The number of characters: The first provisional character number. -Division candidate point: A division candidate point obtained by the vertical distance character number determination unit 17 is adopted. (2) When the first provisional character number <the second provisional character number • The number of characters: The first provisional character number. Division candidate points: From the division candidate points obtained by the vertical distance character number determination unit 17, the number of candidate division points equal to the first provisional character number minus one is adopted in descending order of priority. However, the lower end distance and the upper end distance are very large, the change is very large, and the character height is very small (all are determined by comparison with a preset threshold value). When the number of characters exceeds -1, the number of characters is set to m + 1, and the number of division candidate points is set to m. (3) In the case where the first provisional character number> the second provisional character number: The number of characters: the second provisional character number. -Division candidate point: A division candidate point obtained by the vertical distance character number determination unit 17 is adopted. However, if it is determined that the width of the character delimited at the division candidate point is two characters or more,
The number of characters is increased by setting the x-coordinate of the frame line existing between them as a candidate division point (FIG. 13). This is the end of the description of the operation of the character number determination unit 4.

Next, the noise removing unit 5 performs noise removal on each of the character blocks separated by the character block separating unit 3 after determining the noise (FIG. 14). Finally, in the cutout unit 6, the number of characters determined by the number-of-characters determination unit 4 is determined for each character block remaining without being removed by the noise removal unit 5 using the label separation method and the information on the character height. The character image is cut out by the minute (FIG. 15).

As described above, according to the first embodiment,
Conventionally, even if the number of characters cannot be correctly determined only by the size of the projection because the characters are connected at a plurality of points, the character number determination unit 4 determines the character based on the number of frame lines included in the character block, The number of characters can be determined correctly.

(Embodiment 2) FIG. 16 shows the number-of-characters judging section 4 of the character segmenting apparatus according to the second embodiment of the present invention.
FIG. The configuration / operation other than the character number determination unit 4 is the same as that of the first embodiment. The character number determination unit 4 also includes a front / back lump character number determination unit 51,
The configuration and operation are the same as in the first embodiment, except that the first comparison character number determination unit 18 is changed to the second comparison character number determination unit 52.

The operation of the number-of-characters judging section 4 in the second embodiment will be described below with reference to the flowcharts of FIGS. 16 and 25 showing the operation procedure. Reference numeral 51 denotes a front and rear chunk character number determination unit, and in the character chunk group separated by the character chunk separation unit 3,
Based on the first provisional character number obtained by the frame character number determination unit 14, a third provisional character number is obtained from the positional relationship between the target character block and the preceding and following character blocks (121 in FIG. 25). 52 is the second
, A vertical character number determining unit 17
The actual number of characters in the character block is determined by using the second provisional character number obtained in (1) and the third provisional character number obtained by the preceding and following block character number determination unit 51 (122 in FIG. 25).

The operation of the character number judging unit 4 of the character segmenting apparatus having the above-described structure will be described below with reference to FIG. In the front and rear chunk character number determination unit 51, based on the first provisional number of characters obtained by the frame line character number determination unit 14, in the character chunk group separated by the character number separation unit 3, a frame including the left end of the target character chunk (frame α ) Includes a part of the immediately preceding character block, and if the size of the immediately preceding character block included in the frame α is larger than the size of the target character block included in the frame α, If the number of characters is reduced by one, and the frame including the right end of the target character block (hereinafter referred to as a frame β) includes a part of the character block immediately after, the size of the character block immediately after included in the frame α Is larger than the size of the target character block included in the frame α, the number of characters is reduced by one and the value is set to the third
Is the number of provisional characters.

The second comparison character number determination unit 52 determines the actual character chunk from the second provisional character number obtained by the vertical distance character number determination unit 17 and the third provisional character number obtained by the front and rear chunk character number determination unit 51. The number of characters is determined. As an example of the rule for determining the number of characters, the first provisional number of characters in the rule for determining the number of characters used in the description of the operation of the first comparative character number determination unit 18 in the first embodiment is the third. There is a method in which the determination is made by replacing the number of temporary characters.

As described above, according to the second embodiment,
Based on the principle that only one character exists in one frame, the immediately preceding character block is included in the frame where the left end of the character block exists, or the immediately following character block is included in the frame where the right end of the character block exists. In such a case, it is possible to improve the accuracy of the number-of-characters determination by determining whether the target character block or the character block immediately before or immediately after is a character to be included in the frame.

(Embodiment 3) FIG. 18 shows a noise removing unit 5 of a character cutout device according to a third embodiment of the present invention.
FIG. The configuration and operation other than the noise removal unit 5 may be the same as those of the first or second embodiment, and the operation is the same. FIG. 26 is a flowchart showing an operation procedure according to the third embodiment. FIG.
In FIG. 8, reference numeral 21 denotes a small chunk noise removing unit, which removes small noise from each character chunk separated by the character chunk separating unit 3 (131 in FIG. 26). A character type determination unit 22 determines whether each character block separated by the character block separation unit 3 is print or handwriting. A character type noise removing unit 23 removes the target character block when the target character block is handwritten by the character type determination unit 22 and the character blocks before and after the target character block are printed (FIG. 26, 135).

FIG. 19 is a block diagram of the character type judging section 22. Reference numeral 31 denotes a height determination unit that determines a character type when the target character block is similar in height to the preceding and following character blocks (132 in FIG. 26). Reference numeral 32 denotes a coordinate determination unit, in which the vertical coordinate of the upper end of the target character block is close to the vertical coordinate of the upper end of the preceding or following character block, or the vertical coordinate of the lower end of the target character block is perpendicular to the lower end of the preceding or following character block. When approximating the coordinates,
It is determined that the target character block is a print type (133 in FIG. 26). 3
Reference numeral 3 denotes a handwriting determination unit that determines a character block that is not determined as a print character by the height determination unit 31 or the coordinate determination unit 32 as handwriting (134 in FIG. 26).

The operation of the noise elimination unit 5 of the character segmenting apparatus having the above-described structure will be described below.
The noise removal unit 5 performs noise removal on each character block separated by the character block separation unit 3 after noise determination.
Hereinafter, the detailed operation of the noise removing unit 5 will be described with reference to FIGS. First, the small block noise removing unit 21 removes a character block smaller than a threshold using the size information of each character block separated by the character block separating unit 3 (FIG. 2).
0). Next, in the character type determination unit 22, the character block separation unit 3
The print / handwriting determination of each character block separated by is performed in the following procedure. First, the height determination unit 31 compares the height of the target character block with the height of the preceding and following character blocks, and determines that the target character block is a print type when the heights substantially match (FIG. 2
1). Next, the coordinate determination unit 32 calculates the y coordinate of the upper end of the target character block, the y coordinate of the upper end of the preceding and following character blocks, the y coordinate of the upper end of the target character block, and the y coordinate of the upper end of the preceding and following character blocks. Are compared, and when the y coordinate of the upper end or the lower end substantially coincides, the target character block is determined to be a print character (FIG. 22). The reason why the judgment is made based on the matching of only one of the upper end and the lower end is to make it possible to determine the print type even if a part of the character is missing at the time of binarization by the faint stamp character.
Next, the handwriting determination unit 33 determines that a character block that is not determined as a print character by the height determination unit 31 or the coordinate determination unit 32 is handwritten. The above is the operation in the character type determination unit 22. Next, in the character type noise removing unit 23, the character type determining unit 22
When the target character block is determined by handwriting in, and the character blocks before and after the target character block are determined by print, the target character block is removed (FIG. 23). It should be noted that the number of character chunks before and after used for type determination is not limited to one, and may be changed according to the recognition target image. In addition, it is not necessary that all of the character chunks before and after are type-determined, and may be determined by the ratio or the number of the character chunks whose type is determined.

As described above, according to the third embodiment,
Even for large noise that could not be removed by the conventional method,
When the characters before and after are printed, the noise removing unit 5
Since the character chunk before and after is determined by print type and the target character chunk is determined by handwriting, the target character chunk can be determined as noise and can be removed.

[0034]

As described above, according to the first aspect of the present invention,
According to the character segmentation device described in (1), the number of characters caused by the character number determination method based on only the projection and character height information, which occurs when the characters before and after are linked or when the upper part is open like a handwritten “4”, is eliminated. By taking into account the balance between the existing positions of the entry frame and the characters, the number of characters can be determined more accurately.

According to the character extracting device of the second aspect of the present invention, it is determined whether the target character block or the character block immediately before or immediately after is the character to be included in the frame, and
By correcting the number of characters of the character block based on the result, it is possible to determine the number of characters in consideration of the balance of the number of characters with the surrounding character blocks, and as a result, the accuracy of the determination of the number of characters is improved.

According to the character extracting apparatus of the third aspect of the present invention, it is assumed that a handwritten character is not mixed in a printed character string, and a character / handwriting determination is performed for each character block. If the target character block is handwritten and the character blocks before and after the target are type determination, the target character block is determined to be noise. It can be removed if it is contained within.

According to the character extracting device of the fourth aspect of the present invention, since the height is different from that of the front and rear character judging character blocks, it is possible to perform the front and rear shading type which cannot be determined by the height judgment alone. The character type can be determined only by approximating the vertical coordinates of the character block and the upper or lower end of the character block.

According to the character extracting apparatus of the fifth aspect of the present invention, basically, all the characters in the same field are composed of the same type of font or handwriting arranged regularly. By performing type / handwriting determination using the height / coordinate information of the entire character block, handwriting / type determination can be performed using as much information as possible. A handwritten judgment character block included in the type judgment character block,
That is, more noise can be found and removed.

According to the character extracting method of the sixth aspect of the present invention, the projection and character height information generated when the preceding and following characters are connected or when the upper part is open such as a handwritten "4" Error in the number of characters
By re-evaluating the balance of the existing positions of the entry frame and the character, it is possible to more accurately determine the number of characters.

According to the character extracting method of the present invention, it is determined whether the target character block or the character block immediately before or immediately after is the character to be included in the frame, and
By correcting the number of characters in the character block based on the result, it is possible to determine the number of characters in consideration of the balance of the number of characters with the surrounding character blocks, and as a result, the accuracy of the determination of the number of characters is improved.

According to the character extracting method of the present invention, on the assumption that handwritten characters are not mixed in the type character string, the type / handwriting determination is performed for each character block. If the target character block is handwritten and the character blocks before and after the target are type determination, the target character block is determined to be noise. It can be removed if it is contained within.

According to the recording medium of the ninth aspect of the present invention, the character extracting method of the present invention can be easily introduced into another system, and a system having a function of a recognition unit, or By combining with a recording medium, it is possible to read the characters included in the input pattern only by acquiring the input pattern, and to reduce the burden of confirmation and correction on the user side as a result of the improvement of the recognition performance.

[Brief description of the drawings]

FIG. 1 is a block diagram of a character cutout device according to each embodiment of the present invention.

FIG. 2 is a block diagram of a character number determination unit according to the first embodiment;

FIG. 3 is an explanatory diagram of an example of an input image according to the first embodiment;

FIG. 4 is an explanatory diagram of an example of detection and numbering of vertical frame lines according to the first embodiment.

FIG. 5 is an explanatory diagram of a method of extracting a character block unit according to the first embodiment;

FIG. 6 is an explanatory diagram of a method of storing the number of frame lines / numbers included in a character block according to the first embodiment.

FIG. 7 is an explanatory diagram of a method of obtaining a distance between a selection frame line and a left end of a character block in the first embodiment.

FIG. 8 is an explanatory diagram of a method of obtaining a distance between a selection frame line and a right end of a character block according to the first embodiment.

FIG. 9 is an explanatory diagram of a first provisional character number calculation method based on a positional relationship between a frame line and a character block according to the first embodiment;

FIG. 10 is an explanatory diagram of a distance (upper distance) from the upper end of a circumscribed rectangle to a character in each column in the first embodiment.

FIG. 11 is an explanatory diagram of a distance (lower end distance) from a left end of a circumscribed rectangle to a character in each column according to the first embodiment.

FIG. 12 is an explanatory diagram of a second provisional character number calculation method using the upper end distance and the lower end distance in the first embodiment;

FIG. 13 is an explanatory diagram of a character number correction method according to the first embodiment when the character width is large.

FIG. 14 is an explanatory diagram of noise removal according to the first embodiment.

FIG. 15 is an explanatory diagram of a separation example in units of one character when a character block includes a plurality of characters according to the first embodiment;

FIG. 16 is a block diagram of a character number determination unit according to the second embodiment.

FIG. 17 is an explanatory diagram of correction of a division candidate position (the number of characters) based on a positional relationship with characters before and after according to the second embodiment;

FIG. 18 is a block diagram of a noise removing unit according to the third embodiment;

FIG. 19 is a block diagram of a character type determination unit according to the third embodiment;

FIG. 20 is an explanatory diagram of small noise elimination in the third embodiment.

FIG. 21 is an explanatory diagram of a type determination method based on the height of a character block in the third embodiment.

FIG. 22 is an explanatory diagram of a type determination method based on the degree of coincidence of upper and lower y coordinates corresponding to a missing type according to the third embodiment;

FIG. 23 is an explanatory diagram of a method for removing large noise using a type determination method according to the third embodiment.

FIG. 24 is a flowchart of an operation procedure according to the first embodiment.

FIG. 25 is a flowchart of an operation procedure according to the second embodiment.

FIG. 26 is a flowchart of an operation procedure according to the third embodiment;

FIG. 27 is a block diagram of a character segmentation device in the first conventional example.

FIG. 28 is an explanatory diagram of a character number determination / cutout method in the first conventional example.

FIG. 29 is a block diagram of a character cutout device according to a second conventional example.

FIG. 30 is an explanatory diagram of an example in which the number of characters is determined to be small in the first conventional example.

FIG. 31 is an explanatory diagram of an example of determining a large number of characters in the first conventional example.

FIG. 32 is an explanatory diagram of an example of noise that cannot be removed in the second conventional example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Binarization part 2 Frame line detection part 3 Character block separation part 4 Character number judgment part 5 Noise removal part 6 Cutout part 11 Frame number calculation part 12 Left frame line distance calculation part 13 Right frame line distance calculation part 14 Number of frame line characters Judgment unit 15 Upper end distance calculation unit 16 Lower end distance calculation unit 17 Vertical distance character number judgment unit 18 First comparative character number judgment unit 21 Small block noise elimination unit 22 Character type judgment unit 23 Character type noise elimination unit 31 Height judgment unit 32 Coordinate Determination unit 33 handwriting determination unit 51 front and back mass character number determination unit 52 second comparison character number determination unit 201 image input unit 202 projection unit 203 determination unit 204 deletion unit 205 character cutout unit 206 recognition unit 211 normalization unit 212 blocking unit 213 position Area calculation unit 214 Area calculation unit 215 Noise determination table 216 Noise determination unit 217 Noise removal unit

Claims (9)

[Claims] 1. A binarization unit for binarizing an input image when a slip in which a character group is written in a horizontally continuous non-separable frame is input as an image. A frame line detection unit that detects coordinates of all vertical frame lines included in the binarized image, adds numbers in order, and stores the coordinates and numbers of each frame line; and the binarization unit. A character chunk separating unit that separates the image binarized in each character chunk, a character number determining unit that determines the number of characters included in each character chunk separated by the character chunk separating unit, and the character chunk separating unit. A noise removing unit that removes noise from the character block group separated by the above, and for each character block group remaining without being removed by the noise removing unit, cut out characters by the number of characters determined by the character number determining unit. And a cutout unit to be executed, wherein the character number determination unit is A frame count calculating unit that calculates the number and number of frame lines included in the character chunk from the coordinates of the detected frame line and the coordinates of the character chunk separated by the character chunk separation unit; In the frame line group corresponding to the frame line number calculated by the unit, the frame line closest to the left end of the character block separated by the character block separation unit and the left end of the character block separated by the character block separation unit A left frame line distance calculating unit for calculating a distance, and a frame closest to the right end of the character block separated by the character block separating unit in the group of frame lines corresponding to the frame line number calculated by the frame number calculating unit. A right frame line distance calculation unit that calculates the distance between the line and the right end of the character block separated by the character block separation unit, and the number of frame lines included in the character block determined by the frame number calculation unit,
From the distance between the frame line detected by the left frame line detection unit and the left end of the character block, and the distance between the frame line detected by the right frame line detection unit and the right end of the character block, the first provisional character count of the character block is calculated. A frame line character number judging unit to be judged, an upper end distance calculating unit for calculating a distance from the upper end of the rectangle circumscribing the character chunk separated by the character chunk separating unit to the character chunk for each vertical line, A lower end distance calculation unit for calculating a distance from the lower end of the rectangle circumscribing the separated character lump to the character lump for each vertical line, and a distance from the upper end of the rectangle to the character lump for each vertical line calculated by the upper end distance calculation unit An upper / lower distance character number determining unit for obtaining a second provisional character number of the character block from the distance from the lower end of the rectangle to the character block for each vertical line calculated by the lower end distance calculating unit; and the frame line character number determining unit. The first number of temporary characters and the number of upper and lower distance characters A first comparison character number determination unit that determines the number of characters in the character block using the second provisional character number determined by the determination unit; and the first comparison character number determination unit determines the number of characters determined by the first comparison character number determination unit. Character cutout device to be output. 2. The method according to claim 1, wherein the character number determining unit determines, for each character block separated by the character block separating unit, the frame line character number determining unit using a positional relationship between a target character block and preceding and following character blocks. A mass character judging unit that corrects the first temporary character number to obtain a third temporary character number, and a second temporary character number obtained by the vertical distance character number determining unit instead of the first comparative character number determining unit. And a second comparison character number determination unit that determines the number of characters in the character block by using the third provisional character number obtained by the front and rear lump character number determination unit. The number of characters determined by the second comparison character number determination unit 2. The character segmenting device according to claim 1, wherein the character string is an output of the character number determining unit. 3. A binarizing section for binarizing an input image, a character chunk separating section for separating the image binarized by the binarizing section for each character chunk, and separating by the character chunk separating section. A number-of-characters determining unit that determines the number of characters included in each of the determined character blocks; a noise removing unit that removes a character block determined to be noise from the character block group separated by the character block separating unit; A cutout unit that cuts out characters for each of the character chunk groups left unremoved by the removal unit by the number of characters determined by the character number determination unit, wherein the noise removal unit includes: A small chunk noise removing unit for judging a small character chunk as noise in each of the character chunks separated by the unit, and a character type judgment for performing type / handwriting judgment of each character chunk separated by the character chunk separating unit And a character block targeted by the character type determination unit Handwritten determined, and if the before and after text chunks of target character lumps is print determined in the character type determining unit, character segmentation unit and a character type noise removing unit for removing the object character lumps. 4. In a case where a slip in which a plurality of characters are continuously written in a horizontal direction is imaged and input, the character type determination unit determines that a target character block is similar in height to the preceding and following character blocks. In case,
A height determination unit that determines the target character block as a type, and the vertical coordinate of the upper end of the target character block approximates the vertical coordinate of the upper end of the preceding or following character block, or the vertical coordinate of the lower end of the target character block is When the vertical coordinate of the lower end of the character block is approximated, a coordinate determining unit that determines the target character block as a type, and a character block that is not determined as a type by the height determining unit or the coordinate determining unit is determined as a handwriting. The character segmentation device according to claim 3, further comprising a handwriting determination unit. 5. In the case where a plurality of characters are successively written in a horizontal direction and a slip in which a character entry area is separated in units of fields is input as an image, the character type determination unit may be configured to execute each of the characters in the same field. 5. A field determination unit which determines a character block whose height or vertical coordinate is similar to another character block in the same field as a character block of the same character block as a printed character, and determines that all other blocks are handwritten. Character extraction device according to the description. 6. When a slip in which a character group is written in a non-separable frame is input as an image, the input image is binarized and separated for each character block, and all of the characters included in the binary image are separated. The coordinates of the vertical border are detected and numbered, the coordinates and numbers of each border are stored, and for each character chunk, the coordinates of the detected border and the coordinates of the separated chunk are From the calculated number of frame lines included in the character block, the number of the frame line closest to the left end of the character block and the number of the frame line closest to the right end of the character block are calculated from the calculated frame line numbers. ,
The number of frame lines included in the character block, the distance between the left end coordinate of the character block and the frame line closest to the left end of the character block among the frame lines included in the character block, the right end coordinate of the character block, and the character block The first provisional number of characters in the character block is determined from the distance from the frame line closest to the right end of the character block among the frame lines included therein. For each rectangle circumscribing each character block, The distance to the corresponding character block and the distance from the lower end of the rectangle to the corresponding character block are determined, and the second temporary number of characters in the character block is determined from the magnitude and change of these distances. The actual number of characters in the character block is determined from the number of provisional characters in 2, and the character blocks determined as noise are removed from the character block group. For each character block group remaining without being removed,
A character extraction method that executes character extraction for the determined number of characters. 7. When a slip in which a character group is entered in a non-separable frame is input as an image, the input image is binarized and separated for each character block, and all of the characters included in the binary image are separated. Detects the coordinates of the vertical frame lines and numbers them, stores the coordinates and numbers of each frame line, and for each character block, calculates the coordinates of the detected frame line and the coordinates of the separated character block. , Calculate the number and number of the frame lines included in the character block, and calculate the frame line number closest to the left end of the character block and the frame line number closest to the right end of the character block from the calculated frame line numbers. Asked,
The number of frame lines included in the character block, the distance between the left end coordinate of the character block and the frame line closest to the left end of the character block among the frame lines included in the character block, the right end coordinate of the character block, and the character block The first provisional number of characters in the character block is determined from the distance from the frame line closest to the right end of the character block among the frame lines included therein. For each rectangle circumscribing each character block, The distance to the corresponding character block and the distance from the lower end of the rectangle to the corresponding character block are determined, and the second provisional number of characters in the character block is determined from the magnitude and change of these distances. The third number of provisional characters is determined from the positional relationship of the character blocks, the actual number of characters in the character block is determined from the first provisional character number, the second provisional character number, and the third provisional character number, and noise is determined from the character block group. Removed character chunks, and judge for each character chunk remaining without being removed Character segmentation method for performing a cut-out of character just the number of characters that was. 8. An input image is binarized and separated for each character block, coordinates of all vertical frame lines included in the binary image are detected and numbering is performed, and coordinates of each frame line are determined. After storing the numbers and determining the number of characters for each character chunk, the type and handwriting judgment of each character chunk is performed, and the judgment result of the target character chunk is handwritten, and the character chunk around the target character chunk is determined. If the determination result is a print type, the target character block is determined and removed as noise, and a small character block in the character block group is determined and removed as noise. A character extraction method for extracting characters by the determined number of characters. 9. A recording medium which records a program that realizes the character extracting method according to claim 6 by software.