JPH05114047A - Device for segmenting character - Google Patents

Abstract

PURPOSE:To provide a character segmenting device for segmenting individual character patterns from contact characters coming into contact with each other in an oblique direction or contact characters coming into contact with each other on two positions or more. CONSTITUTION:A character segmenting part 120 segments the character line image of one character from the image data of a document which are quantized to a black-and-white binary value inputted from an image input part 110. A character lump segmenting part 131 in a character segmenting device 130 segments a lump of connected black picture elements from the character line image, a constriction detecting part 132 detects a constricted part of a character line in the lump of the black picture elements. A character pattern segmenting part 133 separates the lump of the black picture elements on the constricted part and segments individual character patterns from contact characters coming into contact with each other in an oblique direction or contact characters coming into contact with each other on two positions or more.

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, and particularly to a contact character which is frequently contacted in an English document or the like and which is contacted in an oblique direction or a contact character contacted at two or more places. The present invention relates to a character slicing device capable of slicing a character pattern.

[0002]

2. Description of the Related Art If character information contained in a document can be accurately obtained from a document image on a medium, an information processing apparatus having various functions, such as an apparatus for storing character information, a character recognition apparatus, and a recognized character can be used. A device that outputs by voice,
It is possible to construct a device that translates recognized characters.

However, in order to realize such an information processing apparatus, character pattern data for each character is accurately cut out from the input character string data obtained from the input character string on the medium, and this is used as a dictionary. Need to match. Therefore, a character slicing device capable of accurately slicing each character pattern data from the input character string data is required.

Conventionally, as an apparatus for cutting out character pattern data for one character from input character string data, the input character string data may be orthogonal to the character string direction (main scanning direction) (hereinafter also referred to as sub-scanning direction). .) Sequentially, a histogram of black bits is created from the number of black bits representing the character line portion on each sub-scan line, and the number of black bits is compared with a predetermined threshold value S. There has been widely known a device for cutting out data of a region having a value of S or more and a value of S or less again as character pattern data.

However, in the above character slicing device, a character which frequently appears in a general English document or the like and in which adjacent characters are in contact with each other as shown in FIG. 4A can be sliced into individual characters. I could not do it.

Therefore, as a character slicing device which can solve such a problem, for example, reference: Japanese Patent Application No. 01-213.
808 was disclosed.

According to the character slicing device of this document, first, a scanning area that is considered to include a delimiter position of each character pattern in the character line direction is set for the character block pattern data, and the area is defined as the character line direction. The second angle
In the second direction from the upper frame and the lower frame of the frame circumscribing the character line portion in the scanning area to the character line portion in the second direction. The maximum distance is obtained, and the position of the maximum distance is set as the position where the character pattern data is cut out.

[0008]

However, in the method of the character slicing device, the contact direction of the character pattern is parallel to the character string direction, and the second scanning direction is the character line direction. This is effective in the case, but in other cases, for example, when two or more character patterns are in contact with each other in an oblique direction as shown in FIG. However, there is a problem that it is not effective at all and it is not possible to properly cut out the character pattern.Therefore, in the above method, the character cutting for correctly cutting out the character pattern from the English document etc. There is a problem that the device cannot be realized and it becomes difficult to realize the above-mentioned various information processing devices.

Further, when this conventional method is applied to a character recognition device, it is difficult to obtain correct character pattern data because the character cannot be cut out correctly, and therefore correct character recognition cannot be performed at all. It was impossible. Further, in order to deal with this problem, a correction work by an operator is required, but this work is complicated, and there is a problem that the processing performance of the character recognition device is remarkably deteriorated and the operability is deteriorated. It has been impossible to provide a character recognition device that is practically satisfactory. The present invention has a problem that, in the character cutting method in the conventional device described above, a character pattern that is in contact with a direction of a specific angle different from the character line direction or the character string direction cannot be correctly cut out. The problem that the character pattern cannot be cut out correctly even if the contact character is touched at two or more places, and the contact character that touches in an oblique direction often appears in documents such as English,
It is possible to accurately extract character information from a document image by appropriately cutting out character pattern data for each character that is touching at more than one place. , Information processing device having various functions described above and highly accurate recognition are possible,
An object of the present invention is to provide a character slicing device capable of constructing a high-performance character recognition device that is easy to operate and can efficiently input characters.

[0010]

In order to solve the above-mentioned problems, the present invention provides a character slicing device for slicing a character pattern from a character line image, and a means for extracting a group of continuous black pixels from the character line image, Constriction detecting means for detecting a narrowed portion in which the line width of the character line forming the black pixel block is smaller than that of the extracted black pixel block, and the black pixel in the detected necked part Is provided with a character pattern cutout unit that cuts out a character pattern by dividing the block into two or more black pixel blocks.

[0011]

[Function] The contact points of contact characters that frequently appear in English documents are
When the line width of a character is sufficiently thin, that is, it is often in a constricted state. Therefore, the constricted part is detected from the series of black pixel blocks cut out from the character line image, and the black pixel blocks are separated in this constricted part and the individual character patterns are cut out so that there are two or more character patterns. It is possible to correctly cut out a character pattern even when the two are in contact with each other in an oblique direction, or when they are in contact with each other at two or more places.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the character slicing device of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram schematically showing the configuration of a character recognition device incorporating the character cutting device of the embodiment.

In FIG. 1, a character recognition device 100 includes an image input unit 110, a character line cutout unit 120, a character cutout device 130 of an embodiment, a character recognition unit 140, and an output terminal 1.
The character segmentation device 130 further includes a character block segmentation unit 131, a constriction detection unit 132, and a character pattern segmentation unit 133.

Characters, figures, symbols, etc. (hereinafter referred to as characters)
The optical signal S from the document in which is described is input to the image input unit 110.

The image input unit 110 photoelectrically converts the optical signal S and quantizes it into an electric signal (for example, a form image, hereinafter) quantized into a black and white binary value in which the character line portion is a black pixel and the background portion is a white pixel. (Referred to as) is generated and the form image data is output to the character line cutout unit 120.

The character line cutout unit 120 cuts out character lines one by one from the form image data input from the image input unit 110 and inputs the character line image data to the character cutout device 130. In the present embodiment, the character lines are cut out by scanning the input form image data with the character line direction as the main scanning direction and the character string direction as the sub-scanning direction to determine the distribution of black pixels in the character string direction. make,
This is performed by cutting out as a character line image 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”.

In the character slicing device 130, first, the character block slicing unit 131 sequentially detects a block of black pixels from the input character line image data, and outputs the block of black pixels to the constriction detecting unit 132 as a character block. To do.

The block of black pixels is detected by the following method.

First, the input character line image data is scanned with the character string direction as the main scanning direction and the character line direction as the sub scanning direction to create a black pixel distribution, and the black pixel distribution is “0. From the position that changes from "1" or more to "1"
From the above, up to the position immediately before changing to "0" is sequentially cut out as a character block.

Next, for each of the cut out character blocks, in order to detect the character gap, the background portion (white bit) at the upper side of the circumscribing frame of the character block is sequentially propagated in the lower side direction, and the background propagation is performed. Create a pattern. This background propagation pattern is defined by the same two-dimensional coordinates as the character block pattern, and for each pixel, a pattern value "1" meaning propagation of the character line portion, and a pattern meaning propagation of the background portion from the upper left direction. The value is “2”, the pattern value is “3” which means the propagation of the background portion from directly above, and the pattern value is “4” which means the propagation of the background portion from the upper right direction.

FIGS. 2A to 2C are diagrams for explaining the method of determining the background propagation pattern value, and the method of creating the background propagation pattern will be described below.

A character block pattern as shown in FIG. 3A is sequentially scanned from the upper side to the lower side with a direction parallel to the character string direction as the main scanning direction, and the pixel value of each pixel is "1". Test whether it is "0". In this test, when the pixel value of the tested pixel is "1", the pattern value "1" meaning the character line portion is stored at the tested pixel position on the background propagation pattern. On the other hand, when the pixel value of the tested pixel is “0”, the pattern value of the tested pixel is determined as follows according to the background propagation pattern value on the scan line immediately before the scan line.

As shown in FIG. 2A, when the background propagation pattern value corresponding to the pixel directly above the pixel to be inspected (indicated by Δ) is 2 to 4 which means a white background, the inspected pixel is inspected. The background propagation pattern value corresponding to the pixel is set to “3”, and FIG.
, The background propagation pattern value corresponding to the upper left pixel of the tested pixel is any of "2" to "4", and
When the background propagation pattern value corresponding to the pixel immediately above is "1",
The background propagation pattern value corresponding to the pixel under test is set to "2", and as shown in FIG. 2C, the background propagation pattern value corresponding to the upper left pixel and the pixel immediately above is "1". And the background propagation pattern value corresponding to the upper right pixel is any of “2” to “4”, the background propagation pattern value corresponding to the tested pixel is set to “4”.

When the above-described processing is performed for each pixel on the uppermost column, the leftmost column, and the rightmost column of the character block circumscribing frame, the character pattern of white pixels is further arranged on the upper column, the left column, and the right column. Columns,
And a background propagation pattern sequence having a pattern value meaning a character background portion. By the above processing, the background propagation pattern as shown in FIG. 3B is obtained for the character block pattern as shown in FIG.

Next, the background propagation pattern corresponding to the lower side of the character block circumscribing frame is examined, and the pattern value surrounded by the background propagation pattern having the pattern value “1” is any value of “2” to “4”. The area that takes is detected. If this area cannot be detected, it is determined that the character block consists of a single character block, and the character block data is output as it is to the constriction detection unit 132. When the area is detected, the area is determined to be the space between the character blocks forming the character block, and the rightmost pixel of the area, that is, the pattern value is changed from a value other than "1" to "1". The X and Y coordinates of the changing pixel (the pixel in the upper left corner is the origin (1, 1) of the coordinates) are stored in the variables X ₁ and Y ₁ as the starting point coordinates of the gap, and the subsequent processing is performed. .. In the background propagation pattern of FIG. 3B, the pattern value on the bottom side is "1111111111".
1133211111111111114333 ", the 14th to 16th" 332 "are surrounded by the pattern value" 1 ", and the pixel address (X ₁ , Y ₁ ) of the pattern value" 2 "at the right end is (16, 31). ) Is stored as the starting point coordinates.

Next, based on the starting point coordinates (X ₁ , Y ₁ ) and the background propagation pattern, the gap coordinates of the character block are sequentially obtained by the following equation (1). However, Ai is the gap coordinate (Xi, Y
i) is the pattern value of the background propagation pattern in i), where i is 1
Take a value in the range of ~ (height of circumscribed frame).

[0027]

[Equation 1]

When a plurality of gaps are detected in the above, processing is performed for each of them to obtain a plurality of gap coordinates.

In the case of the background propagation pattern of FIG. 3B, the gap coordinates are (X, Y) = (16,31), (15,3).
0), (14, 29), ..., (21, 1).

The character block is cut out into two or more character blocks based on the one or more gap coordinates obtained in (3) and each pattern is output to the constriction detection unit 132. In the case of the background propagation pattern of FIG. 3 (B), FIG. 3 (C)
Each pattern divided by the boundary indicated by the broken line is output to the constriction detection unit 132.

Also, from the character block pattern as shown in FIG. 4A, the background propagation pattern as shown in FIG. 4B can be obtained, but the above-mentioned gap is not detected from this pattern,
This pattern is directly output to the constriction detection unit 132.

In the constriction detection unit 132, a character line portion (hereinafter referred to as a constriction) which is thinner than usual is detected from the character block data input from the character block cutout unit 131 in comparison with the width of the character line. , The detected position and the input block of black pixels are output to the character pattern cutout unit 133. The necking is detected by the following method.

First, the input character block data is sequentially scanned to check whether or not each pixel is a black pixel. Next, in the case of a black pixel, scanning is performed in the character string direction, the character row direction, the diagonal left 45 ° direction, and the diagonal right 45 ° direction centering on the pixel, and the black pixel in each direction is scanned. Check the number of consecutive pixels. As a result, when the following expression (2) is satisfied in any one direction, the central pixel is stored in the storage unit (not shown) as a candidate position for the constriction.

[0034]

[Equation 2]

However, K _H , K _V , K _L , and L _R are the character string direction, the character line direction, the left diagonal 45 ° direction, and the right diagonal 45, respectively.
Number of consecutive pixels in the ° direction. T _H , T _V , T _L , and T _R are threshold values in the character string direction, the character line direction, the left diagonal 45 ° direction, and the right diagonal 45 ° direction, respectively. In the present embodiment, T _H , T _V , and
T _L and T _R = 2.

After the detection of the constriction candidate positions is performed for all the pixels of the character block pattern, first, from the obtained constriction candidate positions, five consecutive eight directions around the pixel are white pixels. Points are removed as end points and only the center is left if the constricted candidate positions are continuous,
Then, the constriction position is finally detected.
If no necking position is detected by the above-described processing, the input character block pattern data is directly output to the character pattern cutout unit 133. Figure 4
From the character block pattern as shown in (A), each pixel indicated by “◆” in FIG. 7C is detected as a candidate position for the constriction by the above-described constriction candidate position detection processing. 2A, the two pixels (51, 52) shown in FIG. 5A are detected as the constricted position.

The character pattern cut-out section 133 is a character block pattern (for example, as shown in FIG.
(A)), and the constricted position (for example, 51 in FIG. 5A,
Based on 52), each character pattern is extracted and output to the character recognition unit 140.

The extraction process of the character pattern is performed by replacing the pixel at the constricted position of the character block pattern with a white pixel and then performing the background propagation pattern creation process in the same manner as the character block cutout unit 131. In the case of the character block pattern shown in FIG. 4A, the background propagation pattern shown in FIG. 5B is changed from the character block pattern shown in FIG. 5A obtained by replacing the black pixels at the two constricted positions with white pixels. The pattern boundaries shown by the polygonal lines in FIG. 5C are extracted by sequentially detecting the gap coordinates on the background propagation pattern obtained by the method described above.
By dividing the pattern by this pattern boundary, two character pattern data "f" and "i" are extracted.

The character recognition unit 140 is a sub-pattern extraction unit,
It is composed of a feature extraction unit and a matching unit.

In the sub-pattern extraction unit, the character pattern input from the character pattern cut-out unit 133 is scanned in a plurality of directions, and consecutively over a specific value h (h = 5 in this embodiment) set in advance on each scanning line. Detecting the black pixel row,
By extracting the continuous black pixel row as a black pixel component of the sub-pattern, the sub-pattern for each scanning direction is extracted from the character pattern, and the sub-pattern is sequentially output to the feature extraction unit.

In the present embodiment, the scanning direction is a direction (vertical direction) perpendicular to the character line direction (hereinafter, X-axis direction), a parallel direction (horizontal direction), and 45 ° counterclockwise from the X-axis. The direction (the left diagonal direction) and the direction of 45 ° clockwise (the left diagonal direction) are set, and character patterns are scanned in each of these directions to extract four sub patterns for each direction.

For example, in the extraction of a sub pattern in the vertical direction, the vertical direction is the main scanning direction, the character pattern data is scanned in the vertical direction, and continuous black pixels (black runs) are detected on the scanning line, and L ≧ h. The vertical direction sub-pattern is extracted by extracting the black run of the length L as the black pixel portion of the vertical direction sub-pattern.

Similar to the vertical sub-pattern extraction,
The sub-pattern is extracted when the remaining other direction is the main scanning direction.

In the feature extraction unit, a rectangular area corresponding to the character circumscribing frame of the character pattern is set on the extracted sub-patterns in each direction, and the rectangular area is N × M (N, M).
Is an arbitrary suitable natural number), and the feature quantity representing the length of the character line of each sub-pattern included in each small area is extracted, and the feature quantity is normalized by the size of the character circumscribing frame, Create a feature matrix F consisting of the transformed feature quantity fi,
The feature matrix F is output to the matching unit.

In this embodiment, the number of divisions N and M = 8.
Further, the normalization of the feature amount with the size of the character circumscribing frame is performed with a value of (dX + dY) / 2.
However, dX is the horizontal length of the character circumscribing frame, and dY is the vertical length of the character circumscribing frame. The feature quantity fi is
The number i (i = 1, 2, ..., 1) from 1 to N × M is assigned to each small area.
(M × M) when sequentially expressed in a small area, the number i
Represents the feature amount of the small area and is an element value of the feature matrix F.

The collation unit collates the characteristic matrix F with the characteristic dictionary matrix G of a standard pattern (not shown) prepared in advance, obtains the similarity R represented by the following equation (2), and the similarity R is predetermined. The character name of the dictionary matrix having a value of P or more is set as the candidate character name, and the first candidate character and the second candidate character are ranked in descending order of similarity, and one or more candidate characters are recognized as the recognition result candidate character. Is output from the output terminal 150.

[0047]

[Equation 3]

However, gi represents an element of the dictionary matrix.

Although the present invention has been described in detail above, the present invention is not limited to the above-described embodiments, and various modifications can be made.

For example, the character recognizing unit 110 is not limited to the above-mentioned configuration, and may recognize using various conventionally known methods and systems, and the configuration of the character recognizing unit suitable for the recognizing method. There is no problem as.

Further, in the constriction detecting section of the present embodiment,
Character line width threshold value T when obtaining a candidate position for a waist
(Equation (2)) is a fixed value determined in advance in the present embodiment, but is not limited to this, and the average line width w of character lines from character line data, character block data, etc. A method of calculating according to (4) and determining the threshold value T based on the calculated value may be used. However, in the equation (4), Q is the number of windows in which all 4 pixels in this character pattern are black pixels when the character pattern is scanned with a 2 × 2 window, and A is all the character patterns. The number of black pixels.

W = 1 / (1-Q / A) (4) Further, the image input method, the character line cutout method, the character block cutout method, and the like according to the present embodiment are appropriately various conventionally known methods. May be used.

[0053]

As described above in detail, according to the present invention, a character block cutout section for cutting out a block of black pixels from character line image data and a constriction detection section for detecting a constricted portion of a character line from the character block. And a character cutout part that cuts out a character pattern based on the position detected by the constriction detection part, and the contact position of the contact character that frequently appears in English documents,
Focusing on the fact that when the line width of a character is sufficiently thin, that is, in the majority of cases, the narrowed state is detected, and the narrowed portion is detected from the series of black pixel clusters, and the black pixel is detected in the narrowed portion. By separating the lumps of characters, character patterns are cut out one by one from the contact character, so when two or more character patterns are touching in an oblique direction,
Even when contact is made at more than one place, the character pattern can be cut out properly, and therefore the information of the character pattern can be cut out correctly from the document image, so that the above various information processing apparatuses can be realized. ..

Further, when the present invention is applied to the character recognition device, the character can be cut out correctly, and therefore the character recognition can be performed correctly. Further, since the correction work by the operator is not necessary, it is possible to provide a character recognition device which is easy to operate and which is practically satisfactory.

[Brief description of drawings]

FIG. 1 is a block diagram of a character recognition device incorporating a character cutout device according to an embodiment.

FIG. 2 is a diagram for explaining a method of determining a background propagation pattern value.

FIG. 3 is a diagram for explaining an operation of a character block cutting unit.

FIG. 4 is a diagram for explaining detection of a narrowed candidate position.

FIG. 5 is a diagram for explaining character extraction from an image after detection of a constricted position.

[Explanation of symbols]

 100 character recognition device 110 image input unit 120 character line cutout unit 130 character cutout device 131 character block cutout unit 132 constriction detection unit 133 character pattern cutout unit 140 character recognition unit 150 output terminal

Claims (1)

[Claims] 1. A character segmentation device for segmenting a character pattern from a character line image, comprising means for extracting a continuous block of black pixels from the character line image, and a block of the black pixel formed from the extracted block of black pixels. A constriction detecting means for detecting a narrowed portion in which the line width of the character line is smaller than that of the other portion;
A character cutout device comprising a character pattern cutout means for cutting out a character pattern by dividing into the above-described blocks of black pixels.