JPH06251196A - Character segmenting device - Google Patents

Abstract

PURPOSE:To accurately perform character segmentation and to reduce a time required for data input and a cost by solving a problem in a conventional character segmenting device to divide one character into plural numbers. CONSTITUTION:When a character cluster segmenting part 131 segments a character cluster from character row image data, a character number judging part 132 verifies axial symmetry for the character cluster, and decides whether or not the number of characters in the character cluster is one. When it is decided that the number of characters in the character cluster is not one, the character cluster is further separated to a character pattern by every character at a contact character separation part 133.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character slicing device.

[0002]

2. Description of the Related Art Various information processing devices (for example, a device for storing character information or a device for converting character information into speech) if the character information of the sentence can be accurately obtained from a text image on a medium. Can be realized. In order to realize such an information processing apparatus, it is necessary to accurately cut out each character from the input character string data obtained from the input character string on the medium. However, depending on the input character string, there are characters that are close to each other and characters that are in contact with each other, which makes separation difficult. For example, in Western texts, adjacent characters often come into contact with each other. As a device for separating such contact characters, there is one disclosed in Japanese Patent Application No. 03-275470. In the conventional character slicing device described in this document, attention is paid to the fact that contact characters are often constricted at the contact parts between the characters, and the characters are separated at the constricted parts. explain.

FIG. 2 is a block diagram of a character recognition device using a conventional character slicing device. The character recognition device 20 is
The image input unit 21, the character line cutout unit 22, the character cutout device 23, the character recognition unit 24, and the character code output terminal 25. An optical signal s from a form in which characters, figures, symbols, etc. (hereinafter referred to as characters) are written is input to the image input unit 21. The image input unit 21 photoelectrically converts the input optical signal s into an electric signal (hereinafter referred to as “form image data”) that is binarized into black and white with black pixels in the character portion and white pixels in the background portion, and the form image data Is output to the character line cutout unit 22. The character line cutout unit 22 cuts out character lines one by one from the form image data (hereinafter referred to as character line image data), and outputs the character line image data to the character cutout device 23.
The character cutting device 23 cuts out individual characters from the character line image data (hereinafter referred to as character patterns), and outputs the character patterns to the character recognition unit 24. The character recognition unit 24 recognizes a character from the character pattern, obtains a character code, and outputs the character code to the character code output terminal 25.

The conventional character clipping device 23 will be described below. The conventional character cutout device 23 includes a character block cutout unit 231, a constriction detection unit 232, and a character pattern cutout unit 233. The character block clipping unit 231 scans the form image data in the vertical direction as the main scanning direction (hereinafter also referred to as the y direction) and in the horizontal direction as the sub scanning direction (hereinafter also referred to as the x direction) to distribute the black pixels. To create. Further, from the position where the distribution of the black pixels changes from "0" to "1" or more to the position immediately before the change from "1" or more to "0" is cut out as a character block, and the image data of the character block is extracted. , To the necking detection unit 232. The constriction detection unit 232 detects the constricted portion of the character line from the input image data of the character block, sets the constricted portion as a separation point, and outputs the separation point to the character pattern cutout unit 233. The character pattern cutout unit 233 separates the character blocks at the separation points and outputs the separated character patterns to the character recognition unit 24.

The detection of the constriction of the character line by the constriction detection section 232 is performed as follows. That is, for all black pixels in the extracted character block, the horizontal direction, the vertical direction,
The number of continuous black pixels in the direction of 45 ° to the left and 45 ° to the right is counted, and when any one of the expressions (1) is satisfied, the position of the black pixel is detected as a constricted portion of the character line width. KH <TH KV <TV KL <TL (1) KR <TR where KH, KV, KL, and KR are the number of continuous pixels in the horizontal direction, the vertical direction, the 45 ° left diagonal direction, and the 45 ° right diagonal direction, respectively. Is. Further, TH, TV, TL, and TR are threshold values in the horizontal direction, the vertical direction, the diagonal direction of 45 ° to the left, and the diagonal direction of 45 ° to the right, and are predetermined fixed values.

FIG. 3 shows an example of separating contact characters using a conventional character cutting device. FIG. 3A shows a character 32 described in the form 31, and 33 represents a contact portion between the characters "r" and "s". Figure 3
(B) is the form image data of this character. In the form image data, black pixels are indicated by "0" (black pixels are indicated by "0" in the following form image data). The constriction detection unit 232 selects the portion 3a in FIG. 3B as a separation candidate, the character pattern cutout unit 233 separates the character block at this constricted portion, and outputs the separated character pattern to the character recognition unit 24. .

[0007]

However, in the conventional character slicing device, even if the number of characters in the character block is one,
If the character line width has a constriction, there is a problem that one character is divided into a plurality of parts at the constricted part. For example, in the character block "m" shown in FIG. 4 (a), the conventional character segmentation device finds the constrictions 4a and 4b, and divides "m" into three at these two points. Result of division of "m" Looking only at the part 4c in Fig. 4 (b), it has the same shape as the character of "l (el)", and the operator who corrects the cutting result does not notice the mistake of cutting. , I can't fix it. Further, although an example of separating another character block is given in FIG. 5, for example, in the division example of the character block “W” in FIG. 5C, there is a problem that the 5d part of the division result is mistaken for “V”. was there.

As described above, in the conventional character slicing device, by dividing one character into a plurality of parts,
There was a problem that incorrect data could not be output due to incorrect cutting results. In addition, the operator needs to make corrections to correct the incorrect cutout result.
This work is to find an erroneous cutting result from a huge amount of cutting results, which is complicated and requires time and cost. Therefore, there is a problem in that the original purpose of the character segmentation device, namely, the data input time and the cost required therefor cannot be reduced. Further, when the character slicing device is applied to a character recognizing device, it is necessary to correct an erroneous recognition result, but this work takes time and cost.

According to the present invention, the problem of the conventional character slicing device that one character is divided into a plurality of characters is solved, the character is accurately segmented, and the time and cost required for data input are reduced. Various information processing devices,
For example, it is an object to provide a high-performance character recognition device.

[0010]

SUMMARY OF THE INVENTION In order to solve the above problems, the present invention provides a character block cutout section for cutting out a character block from character line image data and line symmetry of the character block to verify the character block. A character number determination unit that determines whether or not the number of characters in the character block is one character; and if the character number determination unit determines that the number of characters in the character block is not one character It is characterized in that it is provided with a contact character separating section for separating into patterns.

[0011]

6 and 7 are explanatory views of the principle of the character cutting device of the present invention. The principle of the character cutting device of the present invention will be described below. The character slicing device of the present invention, in the character number determination unit, verifies the line symmetry of the character block cut out from the character line image data, thereby determining whether or not the number of characters in the character block is one character, It is characterized in that whether to separate the character block is selected based on the determination result. Thus, the problem of dividing one character of the conventional character slicing device by the character number determination unit is solved,
It realizes accurate character segmentation.

The rationale for using the line symmetry of characters to judge the number of characters is as follows. Considering the case where a Western character that is symmetrical with respect to the central axis of a character contacts another character, the shape of the character block is not symmetrical with respect to the left and right. Some examples are given in FIGS. 6 and 7. For example, in FIG.
In (a), "w" and "i" are in contact. Both “w” and “i” are line-symmetrical to the left and right, but the shapes of the contacted character blocks are not line-symmetrical. In addition, examples of contact characters including line-symmetrical characters are shown in FIG. 6 (b) and subsequent FIGS. 7 (a) to 7 (c), but no character block is line-symmetrical. In addition, Roman characters have left-right line symmetry (i, o,
u, v, M, A, etc.) and a line symmetry (B, C,
D, Q, a, e, f, etc.). From the above, it can be said that the number of characters contained in the character block is 1 if the cut-out character block has line symmetry to the left and right.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1A shows an embodiment of a character recognition device using the character cutting device of the present invention. Character recognition device 10
Is an image input unit 11, a character line cutout unit 12, a character cutout device 13, a character recognition unit 14, and a character code output terminal 1.
It consists of 5. Hereinafter, an example in which the character segmentation device of the present invention is used in a character recognition device will be described.

An optical signal s from a form in which characters are written is input to the image input unit 11. The image input unit 11 uses the optical signal s
Is photoelectrically converted to obtain form image data, and the form image data is output to the character line cutout unit 12. The character line cutout unit 12 cuts out character lines one by one from the form image data, obtains the character line image data, and outputs the character line image data to the character cutout device 13. Character cutting device 13
Then, each character is cut out from the character line image data to obtain a character pattern, and the character pattern is output to the character recognition unit 14. In the character recognition unit 14, from the character pattern,
Obtain the character code, and output the character code to the character code output terminal 1
Output to 5.

Hereinafter, the character extracting device 13 shown in FIG.
Will be described. The character cutout device 13 includes a character block cutout unit 131, a character number determination unit 132, and a contact character separation unit 1.
It consists of 33. Character block image data extraction unit 1
31. The character block cutout unit 131 scans the character line image data in the y direction and the x direction to create a black pixel distribution. Further, from the position where the distribution of the black pixels changes from "0" to "1" or more to the position immediately before the change from "1" or more to "0" is cut out as a character block, and the image data of the character block is extracted. , To the character number determination unit 132. The character number determination unit 132 determines whether or not the shape of the character block is line-symmetrical, and the image data of the character block determined to be line-symmetrical is output to the character recognition unit 14. For a character block whose shape is determined not to be line symmetrical, the image data of the character block is output to the contact character separation unit 133. The contact character separation unit 133 finds the constricted portion of the contact character by the same method as the conventional character cutting device 23, separates the contact character at the constricted portion, and outputs the separated character pattern to the character recognition unit 14.

The character number determination unit 132 determines the line symmetry of the character block as follows. First, the positions of the black pixels corresponding to the “edges” on the left and right of the character block are searched. This process
On the image data of a character block, x is assumed to be a constant y value of the scanning line.
Is to detect the x-coordinate of the position that first appears in the black pixel and the x-coordinate of the position that appears last in the black pixel when scanned in the positive direction. The y value of the scanning line is set to the y of the character block
The minimum value in the direction y = ymin is changed to the maximum value in the y direction y = ymax, and the x coordinate is xl [i], xr [i].
(I = 1, 2, ..., N; n = ymax-ymin + 1 (n corresponds to the height of the character block)) and is stored. next,
The values of w, L, and H in equation (2) are calculated. In equation (2), the value of t is a parameter that can be changed depending on the typeface, etc.
It takes a value in the range of 0 <t ≦ 0.5. w = xmax−xmin L = xmin + t × w (2) H = xmax−t × w where xmin is the minimum value of the character block in the x direction and xmax is the maximum value of the character block in the x direction.

Next, the left edge xl [i] and the right edge xr of the character block are
Let the midpoint of [i] be c [i] (see equation (3)). x =
The number of c [i] between L and x = H is counted (see the equation (4)), and the number is m. When m satisfies the condition of Expression (5), the character is determined to be line symmetric. c [i] = (xl [i] + xr [i]) / 2 (i = 1,2, ..., n) (3) L ≦ c [i] ≦ H (i = 1,2, ..., n) (4) m ≧ a × n (n = ymax−ymin + 1) (5) However, a is a fixed value (0 <a ≦ 1) that takes an appropriate value depending on the typeface of the character.

The above-described processing of the character number determination unit 132 is shown in FIG.
The case of the character block “m” will be specifically described. How to obtain xl [i] and xr [i] in the character block “m” will be described with reference to FIG. 8 in which character patterns are coordinated. Line 74 represents a scan line. When the image data of the character block is scanned in the positive x direction by the scanning line, the first black pixel that appears is 7a and the last black pixel that appears is 7b. in this case,
The y coordinate of the scanning line 74 is the minimum value y in the y direction of the character block.
Black pixel 7a because it is 10 larger than min (y coordinate of line 73)
X-coordinate of the black pixel 7b in the sequence xl [11]
Store in r [11]. The y value of the scanning line is processed by ymin
To the maximum value ymax in the y direction of the character block.

The determination of the symmetry of the character block of FIG. 8 by the character number determination unit 132 is as follows. From FIG. 8, the xmi
The values of n, xmax, ymin, ymax, w, n are: xmin (x value of line 71) = 9, xmax (x value of line 72)
= 40 ymin (y value of line 73) = 17, ymax (y value of line 75)
= 35, w = xmax−xmin = 31 n = ymax−ymin + 1 = 19, so that in equation (2), the value of t is 1/3, and the value of a in equation (5) is 4 If it is set to / 5, the values of L and H will be L = 9 + 31 * 1/3 = 19H = 40-31 * 1/3 = 29. FIG. 9 shows the positions of the character block “m” of x = L and x = H with respect to the image data. In FIG. 9, line x
The value is an L value, and the x value of the line 83 is an H value. X on line 81
The value is xmin, the x value of line 84 is xmax, and the y value of line 85 is ymi.
The y value of n and the line 86 indicates ymax (black pixels corresponding to xl [i] and xr [i] are represented by "#" in FIG. 9). In FIG. 9, n = 19 and m = 19 (that is, in FIG. 9, c [i]
Since all the points of are between the line 82 and the line 83), the formula (5)
Satisfy the condition of.

As described above, since the character block "m" satisfies the condition of the expression (5), the character number determination unit 132 determines that the character block "m" is one line-symmetrical character, and the contact character separation processing is performed. Do not do. Therefore, the character number determination unit 132 outputs the character pattern of the character block “m” to the character recognition unit 14 as it is. Further, in the character block “u” in FIG.
Is 24, the value of n is 24, and the character block “n” in FIG.
Then, in Expression (5), the value of m is 27, the value of n is 27, and FIG.
In the character block “W” of formula (5), the value of m is 26 and the value of n is 2
6 all satisfy the condition of Expression (5). Therefore, the character number determination unit 132 does not perform contact character separation processing on these characters, and the character pattern is used as it is.
Output to.

FIG. 10 shows the result of using the character slicing device of the present invention for a character block of contact characters "p" and "e". In the case of the character pattern shown in FIG. 10, the value of m in expression (5) is 25 and the value of n is 34, which does not satisfy the condition of expression (5). Therefore, the character number determination unit 132 outputs the image data of the character block to the contact character separation unit 133. The contact character separation unit 133 separates the character block at the constricted position 9a,
The separated character pattern is output to the character recognition unit 14.

The operation of the character recognition unit 14 will be described below.
First, the character average line width is calculated from the character pattern. The calculation of the character average line width is performed by calculating the number Q of points at which all the points in the 2 × 2 window are black pixels when the character pattern is scanned through the 2 × 2 window, and all the black pixels in the input character pattern. Count the number A and
This is performed by calculating the character average line width W of the character in the character pattern based on the approximate expression of the character average line width shown in the equation (6). W = A / (A-Q) (6)

Next, four sub-patterns, which are line elements extracted in four directions of horizontal, vertical, left diagonal, and right diagonal, are extracted from the character pattern. In the sub pattern extraction process, for example, in the case of a horizontal sub pattern, the character pattern is scanned in the horizontal direction to detect the succession of black pixels, and the succession number LH of black pixels is calculated by the formula (7).
When the above condition is satisfied, the succession of black pixels is extracted as a horizontal sub-pattern. Similarly, the vertical, left diagonal, and right diagonal sub-patterns are also extracted from the equation (7). RH> 2 x W LV> 2 x W LL> 2 ^1/2 x W (7) LR> 2 ^1/2 x W where LH, LV, LL, and LR are horizontal, vertical, diagonal to the left,
The number of continuous black strokes in the right diagonal direction.

Next, the horizontal, vertical, left diagonal, and right diagonal sub-patterns are divided into small areas, the number of black pixels in each area of each sub-pattern is counted, and from the black pixel meter result and the average line width, Based on (8), the horizontal, vertical, diagonal left and diagonal right feature matrices are extracted. KH (m, n) = BH (m, n) / W KV (m, n) = BV (m, n) / W KL (m, n) = BL (m, n) / W (8) KR ( m, n) = BR (m, n) / W However, KH, KV, KL, and KR are horizontal, vertical, left diagonal,
The right diagonal feature matrix, BH, BV, BL, and BR are horizontal, vertical, left diagonal, and right diagonal black pixel matrices, respectively (m,
In this embodiment, n) is the element number of each matrix. In the present embodiment, the input character pattern circumscribing frame is horizontal,
It is assumed that the image is divided into 25 small areas of 5 × 5, which are created by dividing the area into 5 in the vertical direction.

Then, in the character recognition unit 14, the feature matrix is collated with the feature matrix of the standard characters in the dictionary provided beforehand in the character recognition unit 14 based on the formula (9), and the formula (9) The character code of the standard character having the smallest distance value d is output from the character code output terminal 15 as the recognition result of the input character pattern. d = (Σ (gi-ki) ² ) ^1/2 (9) where gi is an element of the standard character feature matrix and ki is an element of the input character pattern feature matrix.

In the above description, the case where the character slicing device of the present invention is used for Roman characters has been described, but it is clear that this method is effective even for other characters and symbols.

[0027]

As described above in detail, according to the character slicing device of the present invention, the number of characters included in a sliced character block is 1 by utilizing the fact that there are many characters which are line-symmetrical in the Roman alphabet. Whether or not it is a character is determined, and based on the determination, the contact character is separated at the constricted portion of the character line width. Therefore, accurate character cutting can be performed without accidentally dividing one character into a plurality of parts. Further, according to the present invention, since accurate character cutting and data input can be performed, correction work required for data input, time and cost can be reduced. Therefore, various information processing devices can be realized.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a character recognition device incorporating a character clipping device of the present invention.

FIG. 2 is a block diagram showing an embodiment of a character recognition device incorporating a conventional character cutting device.

FIG. 3 is an explanatory diagram of separation of European-language contact characters by a conventional character clipping device.

FIG. 4 is a diagram showing an example of dividing one character by a conventional character cutting device.

FIG. 5 is a diagram showing an example of dividing one character by a conventional character cutting device.

FIG. 6 is a diagram showing an example in which touch characters include left-right symmetrical characters.

FIG. 7 is a diagram showing an example in which touch characters include left-right symmetrical characters.

FIG. 8 is an explanatory diagram of how to obtain a left end xl [i] and a right end xr [i] of a character block.

FIG. 9 is an explanatory diagram of a method of determining the symmetry of a character block.

FIG. 10 is a diagram showing an example of separating contact characters.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 character recognition device 11 image input unit 12 character line cutout unit 13 character cutout device 14 character recognition unit 15 character recognition unit 15 character code output terminal 131 character block cutout unit 132 character number determination unit 133 contact character separation unit

Claims (2)

[Claims] 1. A character block cutout unit for cutting out a character block from character line image data, and a character number determination for verifying line symmetry with respect to the character block and determining whether or not the number of characters in the character block is one character. And a contact character separation unit that further separates the character block into character patterns one by one when the character number determination unit determines that the number of characters in the character block is not one character. Character cutting device. 2. The character slicing device according to claim 1, wherein the line symmetry verification detects a left edge pixel and a right edge pixel from image data of a character block for each scanning line in the sub-scanning direction, A character slicing device characterized by performing distribution of the midpoint positions of the two pixels.