JPH0377183A - Character segmenting device - Google Patents

Abstract

PURPOSE:To improve accuracy for segmenting a character pattern by comparing a block length threshold value with a block length, classifying a black block corresponding to the result of this comparison and selecting the suitable segmenting direction of the character pattern corresponding to this classification. CONSTITUTION:A peripheral distribution preparing means 14 is provided to scan row picture data in a direction almost orthogonal to a direction along a character row, to count the number of black bits on a scanning line for each scanning line and to calculate black bit distribution and a block detecting means 16 is provided to detect an area where the number of the black bits is more than a bit number threshold value. Then, a segmenting direction selecting means 18 is provided to select the direction almost orthogonal to the character row as the segmenting direction when the black block length in the direction along the character row is smaller than the threshold value, and the select a direction obliquely cross the direction along the character row as the segmenting direction when the block length is more than the threshold value, and a segmenting means 20 is provided to segment the character pattern along the selected segmenting direction. Thus, even when plural oblique characters are made close each other, a character segmenting position can be detected for the unit of one character by simple processing.

Description

Detailed Description of the Invention (Industrial Field of Application) The present invention relates to a character cutting device, and particularly to a character cutting device for cutting out character patterns from image data such as a form containing a mixture of characters of a plurality of different types of fonts (for example, italic characters and non-italic characters). This invention relates to a device for accurately cutting out.

(Prior Art) As a conventional technique for cutting out a character pattern (image data of -character character) from image data such as a form, the following, for example, has been proposed.

FIG. 11(A) is a diagram partially showing an example of row image data. In this figure, d indicates a part of the image data for one character line stored in the line buffer, and the white part of this image data d is the character background part, and the symbols 10a to IOe
Each black area shown with an asterisk represents the character line portion of one character. An X-Y orthogonal coordinate system is set on the image data d, with the X-axis direction being along the character lines. In addition, FIG. 11 (8) shows the black bit distribution created by scanning the image data d in the main scanning and sub-scanning directions tY-axis and X-axis directions, and in this figure, the horizontal axis is the X-axis in FIG. The coordinate axis and the vertical axis corresponding to the coordinate axis represent the number of black bits on the scanning line at the coordinate X.

In the first conventional technique, image data d such as a form stored in a line buffer is scanned line-by-line with the main scanning direction being approximately orthogonal to the character line direction, and black bits on the scanning line are scanned for each scanning line. The distribution of black bits is determined by counting the number of black bits, and the cutting position of the character pattern is determined by referring to this distribution.

To determine the cutting position, the number of black bits at each position M@ of the When a position where the number is smaller than the threshold value α is detected, the position immediately before the position is set as the cutting end position×5.

However, for example, as shown in FIGS. 11(A) and 11(B), when italicized characters are very close together, the number of black bits exceeding the threshold α is interrupted character by character, resulting in multiple characters ( If these characters are detected continuously, image data for a plurality of characters will be included between the cutting positions MX and X6, and image data cannot be cut out in units of -characters.

Therefore, Japanese Patent Laid-Open No. 62-127985 proposes a second conventional technique that can extract image data character by character even in such a case.

In this second conventional technique, for example, as shown in FIG. 11(A), image data CI! The image data d is scanned to detect a circumscribing frame G that circumscribes the character line portion of the image data d. Next, circumscribing frame G
Image data dt in the circumscribed frame G is scanned from the upper side to the lower side and from the lower side to the upper side, and a background portion detected by scanning from the upper side and a background portion detected by scanning from the lower side; The image data within the circumscribing frame G is classified into four types of parts: a background part with a vertical line, and a character line part. Next, the image data d within the circumscribed frame G is scanned again to detect the point at which the classification changes (change point), and the cutting position is determined based on the position of this conversion point.

According to this second prior art, italic and non-italic characters are mixed, for example, when only a specific word to be emphasized is printed in italics, and other words that do not need to be emphasized are printed in non-italic fonts. Characters can be extracted character by character from image data.

(Problems to be Solved by the Invention) However, since the above-mentioned second conventional technology has a complicated process, the cutting speed decreases, and the equipment configuration for realizing the above-mentioned process becomes complicated. The problem was that it was difficult to miniaturize.

SUMMARY OF THE INVENTION In order to solve the above-mentioned conventional problems, it is an object of the present invention to provide a character cutting device that can detect the character cutting position for each character by simple processing even when a plurality of italic characters are close to each other.

(Means for Solving the Problem) In order to achieve this object, a character cutting device of the present invention cuts out a character pattern from quantized line image data, and the character cutting device cuts out a character pattern from quantized line image data. a marginal distribution creation means for calculating a black bit distribution by scanning in a direction substantially perpendicular to the scanning direction and counting the number of black bits on each scanning line for each scanning line; A block detection means detects an area as a black block, and when the block length of the black block in the direction along the character line is smaller than the block length threshold, the block detection means selects a direction almost orthogonal to the direction along the character line as the cutting direction, and When the length is equal to or greater than the length threshold, a cutting direction selection means selects a direction that diagonally intersects with the direction along the character line at an angle of 0 degrees (where e is a constant) as the cutting direction; The present invention is characterized by comprising a cutting means for cutting out a character pattern from line image data.

(Function) According to the character cutting device having such a configuration, the block length threshold value and the block length are compared, the black blocks are classified according to the comparison result, and the character pattern cutting direction is determined in an appropriate character pattern corresponding to the classification. Select.

Therefore, the accuracy of cutting out character patterns can be improved.

For example, when a plurality of italic characters are arranged closely as shown in FIG. 11, a black block having a block length exceeding the width of one character is detected. Therefore, assuming that such a situation occurs, the block length threshold is set arbitrarily and suitably, the black blocks are classified into two types: those containing one character and those containing multiple characters, and the black blocks containing multiple characters are When a black block containing characters is detected, it is assumed that a black block containing multiple italic characters arranged in close proximity is detected, and if the cutting direction is set to diagonally intersect with the character line direction at an arbitrary suitable angle, it is assumed that the black blocks are adjacent. To accurately cut out a character pattern for each character from image data of a plurality of italic characters.

(Embodiments) Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that the drawings are only schematically shown to the extent that the present invention can be understood, and therefore the configuration of each component, the flow of input/output signals, the shape, dimensions, and arrangement position M are limited to the illustrated examples. It's not a thing.

FIG. 1 is a diagram for explaining the present invention in detail, and is a functional block diagram schematically showing a configuration example of a character recognition device constructed using a character cutting device M according to an embodiment.

In the figure, 12 is a character cutting device M according to an embodiment of the present invention.
This device 12 scans the line image data in a direction substantially perpendicular to the direction along the character line, and calculates the black bit distribution by counting the number of black bits on each scanning line for each scanning line. a block detection means 16 for detecting an area where the number of black bits is equal to or greater than a bit number threshold with respect to line image data as a black block; Select a direction that is almost orthogonal to the direction along the character line, and when the block length is greater than or equal to the block length value, select a direction that diagonally intersects the direction along the character line at an angle of e degrees (8 is a constant) as the cutting direction. It is characterized by comprising a cutting-out direction selection means 18 for selecting, and a cutting-out means 20 for cutting out a character pattern from line image data along the selected cutting direction.

Reference numeral 22 denotes a character recognition device, and this device 22 includes a preprocessing unit 24 for obtaining fostered line image data of a reading target such as a form, and a recognition target unit from quantized line image data, e.g. A character cutting device 12 that cuts out a character pattern of character-character character, a pattern register 26 that stores the character pattern, and a recognition unit that extracts the feature M of the character corresponding to the character pattern and obtains a character recognition result based on the feature amount. It consists of 28.

Note that the symbol T in FIG. 1 indicates an output terminal connected to an external device (not shown), and the recognition result from the recognition section 28 is inputted to the external device via the output terminal T.

Hereinafter, the character cutting device 12 of this embodiment will be explained in more detail, and at the same time, the character recognition device 22 of the illustrated example will be explained in more detail in order to better understand the present invention.

(Preprocessing Unit) The preprocessing unit 24 in the illustrated example includes a photoelectric conversion unit 24a and a line buffer 24b.

The photoelectric conversion unit 24a detects an area (line area) lFr for one character line of the form to be read based on input format data stored in a format table (not shown), and optically scans this line area. The optical signal S from the form is converted into a quantized electrical signal, and this signal is stored as image data in the line buffer 24b. For example, the image data is black and white 2
It is an electrical signal quantized into values, and the black pixels of the row image data represent the character line portion, and the white pixels represent the character background portion.

FIG. 2 is a diagram showing an example of a form to be read. FIG. 3 is an explanatory diagram of input format data, and FIG. 3 (A) and CB) show examples of input format data and the relationship between input format data and forms.

In the case of the form 30 in the format shown in FIG. 2, for example, as shown in FIG. 3, the X-Y orthogonal The coordinate system is set on the form 30, and the X coordinate Xs and Y coordinate Y3 of the upper left corner point LTI of the line area of the first line as format data, ■ Height h of the line area, ■ Width W of the line head area, ■ Line By using the pitch p and the total number n of lines per document, the position M of the line area on the document can be detected for each character line.

The line buffer 24b stores image data in units of character lines, and has a capacity of, for example, 128×4096 pixels. For example, an X-Y orthogonal coordinate system is set on the line image data, with the tx axis direction along the character line, so that data at pixel positions represented by this coordinate system can be freely read out. .

FIG. 4 shows an example of row image data. In the figure, D indicates image data for one character line, and this line image data D
A part of this is image data d shown in FIG. 11(A). The X-axis and Y-axis in FIG. 4 correspond to the X-axis and Y-axis in FIG. 11(A).

(Character segmentation device) a: Peripheral distribution creation means The peripheral distribution creation means 14 of this embodiment scans the line image data line-by-line in a direction substantially orthogonal to the direction along the character line as the main scanning direction, and The number of black bits on a scanning line is counted for each scanning position, and a distribution of the number of black bits in the column direction is created.

The distribution of the number of black bits of the row image data D shown in FIG.
As shown in the figure.

b Ni block detection means The block detection means 16 of this embodiment refers to the distribution of the number of black bits created by the marginal distribution creation means 14, and detects that the distribution of the number of black bits is 1 or more, for example, by setting the bit number threshold THL tTHL = 1. The area is detected as a black block.

And the starting end value M M of each black block in the character line direction
S and terminal position purchase Me'8 are detected. In this example, when the distribution of the number of black bits changes from ○ to 1 or more, the X coordinate at which the distribution of the number of black bits becomes 1 or more is the starting value 11M5, and when the number of black bits changes from 1 or more to ○. The X coordinate where the distribution of the number of black bits is 1 or more is detected as the end position M.

FIG. 5 shows the detected starting position M and terminal position WM+-
An example was shown.

C: Cutting direction selection means The cutting direction selection means 1 of this embodiment determines the length (block length) P of each black block in the direction along the character line based on the starting end positions M8 and M6 detected by the block detection means 16. Calculate. Block length P is P = ME - MS
It can be expressed as +1.

Then, set the block length P of each black block to the block length value 2
(for example, Z=35).

The cutting direction selection means 18 outputs a cutting direction selection signal C according to the result of comparing the block length P and the threshold value 2. p
When <z, a selection signal indicating that a direction substantially perpendicular to the direction along the character line is selected as the cutting direction; for example, C.
=O, and when P≧C, outputs a selection signal, e.g., C=1, indicating that a direction that diagonally intersects with the direction along the character line at an angle of e degrees (where e is a constant) is selected as the cutting direction. A 0 selection signal C is output for each black block.

In this example, when P<Z, the black block with the block length pt may contain one character, and when P≧Z, the black block with the block length P may contain multiple characters. It is considered as a certain black block.

d: Cutting-out means The cutting-out means 20 of this embodiment includes the cutting-out direction selection means 1
A selection signal C is input for each black block from 8, and a character pattern is cut out from the line image data along a cutting direction corresponding to this selection signal C.

i) When P<Z, when the selection signal C of the row image data represents p<zv, the black block B from which the black block length P has been obtained is considered to be a block of characters that are not italicized, and this black block B, A character pattern is cut out based on the starting end value MMs and the ending end position MM.

For example, the black block with the alpha head '○' shown in FIG. 4 is the black block B.

In this case, the cutting means 20 is parallel to the selected cutting direction, that is, parallel to a direction substantially perpendicular to the direction along the character lines (in the illustrated example, the Y-axis direction), and is parallel to the direction along the character lines. M
Cutting line K + s and + passing through s-8 and Mt+tt
Set +- (where s and t are zero or any suitable positive integer value). FIG. 4 shows examples of cutting lines KI5 and KIE with dotted lines.

And +8 to these cutting lines. The area between the KIEs is set as a cutout area, and the line image data within this cutout area is cut out as image data for one character, that is, a character pattern. The character pattern is cut out character by character and stored in the pattern register 26.

A direction almost orthogonal to the character line of the cutout area (the upper limit position and lower limit position in this direction can be obtained from format data, for example,
The upper side position and the lower side position may be set in a direction substantially perpendicular to the character line of the line area. In the illustrated example, these upper side positions and lower side positions can be expressed by Y coordinates. In addition, the character line portion in the line image data (in a direction almost perpendicular to the character line of the line circumscribing frame 9 that circumscribes this (upper position 119 T and lower position 11
911 may be set as the upper limit position and lower limit position of the cutting area.

ii) When P≧Z, the row image data selection signal C′tJ<1221, the black block B2 that obtained the black block length P is considered to be a black block that may contain multiple italic characters. . For example, the black block of the word LINIC8' shown in FIG. 4 is black block B2.

In this case, the cutting means 20 sets the scanning area based on the start position Ms and ME of the black block B2. The scanning area Iu is, for example, an area of row image data between WMs-p and ME+q (where p and q are zero or any suitable positive integer value).

Then, the line image data in the scanning area is scanned in the cutting direction selected at this time, that is, in the direction that diagonally intersects with the direction along the character line at an angle of 0 degrees, and from the self-scanning line to the black scanning line. The position MC of the black scanning line when changing
II and the position Cw of the black scanning line when it changes from the black scanning line to the self-scanning line. Here, a scanning line in which there is no black pixel representing a character line portion on the scanning line is referred to as a self-scanning line, and a scanning line in which there is a black pixel representing a character line portion on the scanning line is referred to as a black scanning line. The positions HCe and Cw can be represented, for example, by the position where the black scanning line intersects the x-axis.

e (This can be set arbitrarily and suitably depending on the font of the character to be read, but in this example, it is set to 0275 degrees, for example.

Next, the cutting means 20 sets 21i and K2E% to the cutting line based on the positions MCB and Cw.

In this case, the cutting means 20 is parallel to the selected cutting direction, that is, parallel to the character line (in the illustrated example, the X-axis direction), and at the positions Ms-v and ME.
+, set 2S and 2E to the cutting line passing through W (where ■, W is zero or any suitable positive integer @). In FIG. 4, examples of 2S and 2E are shown with dotted lines.

Then, the area between 2S, K, and 2E on these cutting lines is set as a cutting area, and the line image data within this cutting area is cut out as one character's worth of image data, that is, a character pattern. The character pattern is cut out character by character and stored in the pattern register 26.

The upper limit position and lower limit position in the direction substantially perpendicular to the character line of the cutout area may be set in the same manner as in the case of P<Z described above.

Figure 6 shows an example of a character pattern stored in the pattern register. As shown in the figure, an X-Y orthogonal coordinate system with an arbitrary suitable position N as the origin is set on character pattern b. , character pattern data at pixel positions represented by this coordinate system can be freely read out from the pattern register 26 (this is not possible).

The pattern register 26 has a capacity of, for example, 128 x 128 pixels.
It has l.

According to the character cutting device 12 of this embodiment, the processing up to the detection of the cutting area is very simple, so there is an advantage that the cutting speed is fast. Furthermore, since the processing is simple, there is an advantage that the device configuration can be simplified and the device can be made smaller.

(Recognition unit) The recognition unit 2 scans the character pattern stored in the pattern register 26, extracts and recognizes the feature amount of the character pattern, and outputs the recognition result of the character pattern, for example, the character code specified in the JIS standard. .

Although the configuration of the recognition unit 28 can be various conventionally known configurations, in this embodiment, the recognition unit 28 is a recognition unit having the following functions.

The recognition unit 28 of this embodiment first detects a character circumscribing frame circumscribing a character line portion of a character pattern.

Roughly calculate the line width W for the character pattern. The line width W is expressed by the following conventionally known approximation formula.

W= 1/ (1-(Q/A)) However, Q is the number of states in which all points within the window are black bits when a character pattern is scanned using, for example, a 2 x 2 window, and A is the number of states in which all points within the window are black bits. Represents the total number of black bits of the character pattern within the character circumscribing frame.

Threshold value βth used to extract a sub-pattern based on this line width W=N-WL (where N is an arbitrary constant value set for each main scanning direction when extracting a sub-pattern).

Then, the character pattern is scanned in multiple directions, and for each scan, the number of consecutive black bits on the scanning line is A! Detect, threshold value u t
Based on h and the number of consecutive black bits (1), sub-patterns for each scanning direction are extracted from the character patterns. A continuous number of black bits such that l≧A th is extracted as the black bit portion of the subpattern, and white bits on the scanning line and a black bit group such that l<ltl are extracted as the white bit portion of the subpattern.

Next, the recognition unit 28 divides the area of the sub-pattern corresponding to the area within the character circumscribing frame of the character pattern into NXM (N and M are natural numbers) divided areas, and calculates the character line length within the divided area for each divided area. , and normalize the extracted feature values of each divided region according to the size of the characters. For example, using (ΔX+ΔY)/27a as the character size,
The feature JIV is normalized by dividing the feature amount of each divided region by the size of the character. However, ΔX is the length of the character circumscribing frame on the character pattern in the direction along the X-axis, and ΔY is the length of the character circumscribing frame on the character pattern in the direction along the Y-axis.

Then, the recognition unit 28 obtains a feature matrix for each character from the characteristic quantities obtained for each divided region of each sub-pattern.

Figures 7 (A) to (D) show feature matrices for each scanning sub-pattern when the character pattern of the character "U°" shown in Figure 6 is scanned in four different directions to extract four sub-patterns.
Shown below.

Next, the feature matrix of the character pattern is compared with the dictionary matrix, and the character name corresponding to the dictionary matrix with the largest similarity R is output as the recognition result, where i is the number assigned to each divided area, and f is the number i. The normalized feature amount of the divided region marked with and 91 represent the feature ffi:l of the dictionary matrix corresponding to the feature amount fi.

The recognition unit 28 of this embodiment has a plurality of, for example, two types of dictionary matrices for one category. 2f4 prepared for one category, for example the character 'su'! The dictionary matrices of are shown in FIGS. 8 and 9. Figure 8 (A)
The dictionary matrices shown in ~(D) are shown in Figures 7(A)~(D).
The dictionary matrices in italics for each sub-pattern corresponding to , and the dictionary matrices shown in FIGS. 9(A) to (D) are standard dictionaries for each sub-pattern corresponding to FIGS. 7(A) to (D). It is a matrix. The similarity R between the feature matrix in Figure 7 and the dictionary matrix in Figure 8 is R = 11.6.
, and the degree of similarity between the feature matrix in FIG. 7 and the dictionary matrix in FIG. 9 was circle=1.0.

FIG. 10 shows an output example of the recognition result of characters in the line image data shown in FIG. 4. Incidentally, the output example shown in FIG. 10 is a result of the recognition, for example, a character corresponding to a character code is printed.

This invention is applicable to the embodiments described above (but is not limited thereto, and therefore includes the configuration, operation, flow of operation, input/output signals, flow of input/output signals, numerical conditions, shapes, dimensions, and arrangement positions of each component). can be changed as desired.

(Effects of the Invention) As is clear from the above description, the character segmentation device of the present invention compares the block length threshold and the block length, classifies the black block according to the comparison result, and classifies the black block according to the classification. Select the appropriate character pattern cutting direction according to the desired character pattern. Therefore, the accuracy of cutting out character patterns can be improved.

[Brief explanation of drawings]

Fig. 1 is a functional block diagram providing a detailed explanation of the present invention, Fig. 2 is a diagram showing an example of a form to be read, and Fig. 3 (
A) and (B) are diagrams for explaining format data, Figure 4 is a diagram showing an example of line image data, Figure 5 is a diagram showing an example of black bit number distribution, and Figure 6 is a diagram showing an example of the black bit number distribution. Figures showing examples of character patterns; Figures 7(A) to (D) are diagrams showing examples of character pattern feature matrices; Figures 8(A) to (D) and Figures 9(A) to (D). ) is a diagram showing an example of different types of dictionary matrices prepared for one category, Figure 10 is a diagram showing an example of recognition result output, and Figure 11 (A) and CB) are row image data and black FIG. 3 is a diagram showing an example of bit distribution numbers. 12... Character cutting device, 14... Surrounding distribution creation means 16... Block detection means 8... Cutting direction selection means 20... Cutting out means.

Claims (1)

[Claims] (1) In a character cutting device that cuts out a character pattern from quantized line image data, the line image data is scanned in a direction substantially orthogonal to the direction along the character line, and black bits on the scanning line are scanned for each scanning line. a marginal distribution creating means for calculating a black bit distribution by counting the number of black bits; a block detecting means for detecting as a black block an area in which the number of black bits is equal to or greater than a bit number threshold with respect to the line image data; When the block length of the black block is smaller than the block length threshold, a direction substantially perpendicular to the direction along the character line is selected as the cutting direction, and when the block length is greater than or equal to the block length threshold, the direction along the character line is selected as the cutting direction. and a cutting direction selection means for selecting a direction that diagonally intersects with the character pattern at an angle θ degrees (where θ is a constant), and a cutting means for cutting out a character pattern from the line image data along the selected cutting direction. A character cutting device featuring: