JPH04338889A - Method and device for character recognition - Google Patents

Abstract

PURPOSE:To accurately recognize even an inclined character like a character, which is manually stuck to a drawing or the like, in a short time by making multiple templates of a dictionary and giving priority levels to respective templates. CONSTITUTION:A dictionary 11 is made of multiple templates, and a template of erect characters and templates of variously inclined characters are registered in the dictionary 11, and priority levels are given to these templates in the order of possibility of matching to characters on the surface of paper. Meanwhile, characters written on the surface of paper like a drawing are read in by a scanner 1, a frame memory 9, a character pattern memory 8, a character area segmenting part 6, etc., to obtain a character picture. This character picture is successively collated with plural templates in the order of priority level by a character recognizing part 7, and characters of the template best matched to this character picture are recognized as characters on the surface of paper.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition system, and more particularly to a system for recognizing characters and symbols in drawings when the drawings are automatically input into a computer system.

0002

2. Description of the Related Art In recent years, CAD/CAM systems that utilize computers have become widespread in the design and manufacturing departments. Along with this, there has been an increasing need to automate the input of drawings as a means of converting drawings, mechanical drawings, etc. written on paper into CAD data that can be processed by a computer. Recognition of characters and symbols is important when inputting drawings. Conventionally, drawings are input as follows. (1) Read the drawing with a scanner, etc., binarize it, and create raster data. (2) Extract character and symbol areas (hereinafter referred to as character areas) from raster data based on size. Then, assuming a rectangle that circumscribes the character area and whose scanning direction is horizontal, character area data consisting of the coordinates of the upper left corner and the lower right corner of the circumscribed rectangle is created. FIG. 8 shows an example of a character area. In FIG. 8, reference numerals 21 and 22 indicate circumscribed rectangles indicating graphic areas, and 23, 24, 25, and 26 indicate circumscribed rectangles indicating character areas. Each rectangular area is represented by the coordinates (X, Y) of the upper left corner and lower right corner (indicated by dots).

(3) For all character area data, create character string data that defines a character string that integrates adjacent character areas within a predetermined character interval on a straight line with a certain arbitrary slope. The character string data includes the width W and height H of a rectangle circumscribing the entire character area included in the character string, the coordinates (X, Y) of the upper left corner of the rectangle, and the center of the first character area and the last character area. It consists of an angle θ formed by a line connecting the centers and the scanning direction. However, the angle θ of a character string composed of one character is a predetermined value. FIG. 9 shows an example of a character string. In FIG. 9, 27 is the character area 24, 2
5 and 26. The coordinates (X2, Y2), width W2, height H2, and angle θ2 of point 27a at the upper left corner of rectangle 27 represent the character string data of this character string. 23 is a character string composed of one character, and the character string data is the coordinates (X1,
Y1), width W1, height H1, and angle θ1. In this case, θ1=0.

(4) The following (5
) to (6) are performed. (5) For each character, refer to the character area data, cut out the character area from the raster data, and calculate the angle of the character string (
θ2 in FIG. 9) Rotate it to an upright position. Processing to match the size of this image with the template size (normalization)
is applied to create a character image of that character. (6) Recognize character images as follows. The character image is compared with the template in the dictionary, and the code of the template with the smallest distance is taken as the recognition result. However, if the distance is large, there is a high possibility of misrecognition, so a maximum allowable degree of discrepancy is set as a threshold for determining whether the recognition was successful or not. , it is assumed that it could not be recognized (rejected).

[0005]

[Problem to be solved by the invention] When creating new maps, wiring diagrams, illustrations, etc., printed characters have a higher recognition rate than handwritten characters. Sometimes a method is used in which a piece of paper with the necessary characters or symbols printed on it is pasted onto the drawing. When cutting and pasting printed characters manually like this,
It is difficult to paste the letters without tilting them. On the other hand, as described above, in conventional drawing recognition in automatic drawing input, for character strings made up of a plurality of characters, the inclination of the characters is determined from the connection of the characters. However, in the case of a character string consisting of one character or a symbol that often exists alone, the angle cannot be specified, so the angle of the single character string is determined in advance.

However, as mentioned above, in the case of manual cutting and pasting work, it is difficult to accurately paste prescribed characters at a prescribed angle, and it is natural for the letters to be slanted to some extent. Therefore, in the automatic input procedure (5), even if a character image is obtained by cutting out a character area and rotating it into an upright position, the character image may be somewhat tilted and cannot be recognized. The present invention was made in view of the problems in the prior art described above, and it is an object of the present invention to provide a character recognition method and device that can accurately and quickly recognize even slightly slanted characters and symbols. shall be.

[0007]

[Means for Solving the Problems] A character recognition method according to the present invention sequentially matches an image of a character or symbol with a plurality of templates each representing various postures of the character or symbol, and selects a template that best matches the image. Configured to recognize characters or symbols. The above matching is performed against multiple templates.
The larger the inclination angle of the character or symbol from the upright state, the lower the priority is given, and the processing is performed according to this priority.

The character recognition device according to the present invention includes a means for reading characters or symbols drawn on paper to form a character image, and a means for reading characters or symbols drawn on paper to form a character image, and a means for reading characters or symbols drawn on paper to form a character image; , and a character recognition unit that recognizes characters or symbols by sequentially comparing character images with the plurality of templates in the dictionary. The character recognition unit gives the highest priority to the template representing the upright state of the character or symbol, and thereafter gives lower priority to the larger the inclination angle, and sequentially processes the character image and the plurality of templates according to this priority. A matching method is used to extract the template that is closest to the character image and within the tolerance range, recognize the characters or symbols in this template as characters or symbols on the paper, and extract the template that is closest to the character image and within the tolerance range. It is configured to include means for rejecting when it does not exist.

[0009]

[Operation] With the above configuration, the dictionary is made into a multi-template, and templates for upright characters and templates for characters with various inclinations are registered in the dictionary. The plurality of templates described above are prioritized in the order of their likelihood of matching with the characters on the paper. Read the characters drawn on the paper such as a drawing, obtain a character image, compare this character image with the plurality of templates listed above in order of priority, and select the characters from the most matching template with the characters on the paper. Certify. This makes it possible to recognize characters that are slightly tilted, and also prevents erroneous recognition. Also,
Recognition time can be shortened.

0010

Embodiments Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram of an embodiment of a character/figure recognition device according to the present invention. In FIG. 1, a scanner 1 scans a paper surface such as a drawing and converts characters, symbols, figures, etc. into binary raster data. At this time, the scanner control section 2 is in charge of controlling the scanner 1. The scanner control unit 2 stores the read raster data in the frame memory 9. The character/figure separation unit 3 is a raster/graphic separation unit 3.
The sizes of connected black pixel components are measured from the data, and those that fit a predetermined character size are selected as character candidates. Use objects other than character candidates as graphic candidates. For character candidates,
Character area data as described above is created and stored in the character/figure memory 8. The vector generation unit 4 performs raster-vector conversion (broken line approximation) on the graphic candidate and stores the generated vector data in the character/graphic memory 8 .

The character string extraction section 5 refers to the character area data stored in the character/graphic memory 8, creates character string data, and stores it in the character/graphic memory 8. The character area cutting unit 6 refers to the character area data and character string data stored in the character/graphic memory 8, and extracts the character area data from the frame memory 9.
A character area image is cut out from the raster data stored in , rotated, and normalized to create a character image. The character recognition unit 7 performs character recognition by referring to the character image and a dictionary 11, which will be described in detail later, and stores the character recognition result in the code memory 10. The communication control unit 12 outputs vector data stored in the character/figure memory 8 and character data stored in the code memory 10 to the CAD system.

The dictionary 11 is constructed as follows. (1) Make the dictionary 11 a multi-template. In other words, images of all characters or symbols (hereinafter, characters and symbols are collectively referred to as characters) in an upright state and images with a certain inclination are collected as a template group, and as a set of these template groups. A dictionary 11 is configured. For characters that may be tilted in both left and right directions, separate templates are provided for characters tilted to the left and to the right. (2) Prioritize each template group. Add the highest priority to the template group created with upright images,
The next highest priority is given to the template group with the angle that is most likely to be tilted. In this way, priority is added to all template groups. In FIG. 1, the dictionaries 11 are each m
A group of n templates T consisting of character templates
G1, TG2...TGn. The m character templates of the template group TG1 with priority level 1 are all erect character templates. Template group T with priority level 2
G2 has a template of m characters slightly tilted from the upright characters. Thereafter, as the priority of the template group decreases, the slope of the m characters increases.

The character recognition operation of the character/figure recognition device shown in FIG. 1 will be explained below with reference to the flowchart shown in FIG. (1) The drawing is read by the scanner 1 and the raster data is stored in the frame memory 9 (S1). (2) Create character area data from the raster data and store it in the character/figure memory 8 (S2). (3) The character string extraction unit 5 creates character string data and stores it in the character/graphic memory 8 (S3). (4) Perform the following processing on all characters in the string. (5) The character area cutting unit 6 refers to the character area data stored in the character/graphic memory 8, cuts out a character area from the raster data stored in the frame memory 9 (S4), and extracts a character area from the raster data stored in the frame memory 9. The image is rotated by the angle θ to restore the upright state (S5), and this image is normalized (S6).
), character image (S7). In addition, the above (1) to (5)
) are steps (1) to (5) in the prior art described above.
is the same as

Next, the character recognition unit 7 recognizes the character image through steps (6) to (9) below. (6) Set the priority to the highest value (S8). (7) Compare the character image with all templates having the priority (S9). If the distance to all templates is greater than the maximum permissible degree of mismatch (S10, S11), it is assumed that the template cannot be recognized with the priority. If not, it is assumed that recognition has been achieved, and the code Tmin of the template with the smallest distance is set as the recognition result (S12), and the process ends. (8) The priority is lowered by one rank and the process of (7) is performed until the character image can be recognized (S14, S15). (9) If no recognition result is obtained by comparing with all the priorities, it is rejected (S14, S16).

A specific example will be explained below. Now, it is assumed that a recognition process for drawing 14 shown in FIG. 3 is performed. Drawing 1
4, symbols 15, 16 and 17 and some straight lines are drawn. Symbols 15, 16, and 17 are to be recognized. Symbol 15 is a 4 drawn exactly horizontally.
Square shape, symbol 16 is a circle slightly tilted to the right, symbol 17
is a triangle tilted slightly to the left. Here, the angle of one character string is 0 degrees. On the other hand, the dictionary 11 is
and three template groups TG1, TG2, shown in 6.
Consists of TG3. Template group TG1, TG2, T
G3 is given priority 1, 2, and 3, respectively, and priority 1
, 2, and 3 have higher priorities in this order. The template group TG1 with priority level 1 consists of erect symbol templates A1, B1, C1, D1...X1.

Template group TG2 with priority level 2 includes templates A21 and B that include symbols slightly tilted to the left.
21, C21, D21...
...X22 template group TG22. Template group TG3 with priority level 3 includes templates A31, B31, and C31 that include symbols further tilted to the left.
, D31...X31, a template group TG31,
The template group TG32 includes templates A32, B32, C32, D32, . . .

The operation of recognizing each symbol 15, 16 and 17 is performed as follows. (1) Recognition of symbol 15 The character area of symbol 15 is cut out and normalized to create a character image. The character image 18, which looks like 18 in FIG. 7, is compared with the template group TG1 of priority 1. In template group TG1 with priority 1, the distance from template A1 is the minimum and the value does not exceed the maximum allowable degree of mismatch, so the recognition result is set as a square and recognition is terminated. (2) Recognition of symbol 16 The character area of symbol 16 is cut out and normalized to create a character image. The character image looks like 19 in FIG. The character image 19 is compared with the template group TG1 of priority 1. In template group TG1 with priority 1, the distance from template D1 is the minimum, but since the minimum distance exceeds the maximum allowable degree of mismatch, it cannot be recognized. In the conventional technology, it ends here, so symbol 16
cannot be recognized. Next, the character image 19 is compared with the template group TG2 of priority level 2. In template group TG2 with priority level 2, the distance from template D22 is the minimum and the value does not exceed the maximum allowable mismatch degree, so the recognition result is set as a circle and recognition is terminated.

(3) Recognition of symbol 17 The character area of symbol 17 is cut out and normalized to create a character image. The character image looks like 20 in FIG. The character image 20 is compared with the template group TG1 of priority 1. In template group TG1 with priority 1, the distance from template B1 is the minimum, but since the value exceeds the maximum allowable degree of mismatch,
This means that it could not be recognized with priority level 1. In the conventional technique, symbol 17 cannot be recognized because the process ends here. Next, the character image 20 is transferred to the template group T with priority level 2.
Check with G2. In template group TG2 with priority level 2, the distance from template B21 is the minimum, but since the value exceeds the maximum allowable degree of inconsistency, it means that it could not be recognized with priority level 2. Next, the character image 20 is compared with the template group TG3 of priority level 3. In template group TG3 with priority level 3, the distance from template B31 is the minimum and the value does not exceed the maximum allowable degree of mismatch, so the recognition result is set as a triangle and recognition is terminated.

[0019]

[Effects of the Invention] According to the character recognition method according to the present invention, the dictionary is made up of multiple templates, each template is given a priority order, and characters are recognized by comparing character images with a plurality of templates according to the priority order. As a result, even slanted characters that are pasted manually on drawings or the like can be recognized accurately and in a short time. Therefore, it greatly contributes to improving the recognition ability of character/figure recognition devices.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the configuration of the present invention.

FIG. 2 is a flowchart showing the operation of the system of FIG. 1;

FIG. 3 is a drawing to be recognized in an embodiment of the present invention.

FIG. 4 is a diagram showing a template group with priority level 1.

FIG. 5 is a diagram showing a template group with priority level 2.

FIG. 6 is a diagram showing a template group with priority level 3.

FIG. 7 is a diagram showing a character image cut out from the drawing of FIG. 3;

FIG. 8 is a diagram for explaining a character area.

FIG. 9 is a diagram for explaining character string data.

[Explanation of symbols]

1 Scanner 2 Scanner control section 3 Character/figure separation section 4 Vector generation section 5 Character string extraction section 6 Character area extraction section 7 Character recognition section 8 Character/figure memory 9 Frame memory 10 Code memory 11 Dictionary 12 Communication control section 14 Drawing 15 , 16, 17 Symbols 18, 19, 20 Character images 21, 22 Graphic areas 23, 24, 25, 26 Character areas 27
Character strings TG1 to TGn, TG21, TG22, TG31, TG
32 Template groups A1-X1, A21-X21, A22-X22, A31
~X31, A32~X32 template θ1,
θ2, W1, W2, H1, H2 Character string data S
1~S15 steps

Claims (2)

[Claims] [Claim 1] A plurality of templates each representing characters or symbols having different inclination angles are prepared,
A lower priority is given to the plurality of templates as the inclination angle increases, an image of a character or symbol is sequentially compared with the plurality of templates according to the priority, and the character or symbol is selected by the template that best matches the image. Or a character recognition method characterized by recognizing symbols. 2. A means for reading characters or symbols drawn on paper to form a character image; and a plurality of means for reading characters or symbols drawn on paper to form a character image; The highest priority is given to a dictionary having a template and a template representing an upright state of the character or symbol, and a lower priority is given as the inclination angle increases, and according to this priority, the character image and the plurality of characters are The templates are sequentially compared, the template that is closest to the character image and within the tolerance is extracted, and the characters or symbols of this template are recognized as the characters or symbols on the paper, and the template is extracted that is closest to the character image and within the tolerance. 1. A character recognition device comprising: a character recognition unit that rejects a character when a certain template does not exist.