JPH01255986A - Preparation of multi-font dictionary - Google Patents

Abstract

PURPOSE:To prepare a single dictionary at a high recognition ratio for the multiple types of character fonts by averaging the feature quantities extracted from the plural character fonts, and preparing the dictionary to recognize types. CONSTITUTION:An inputted document is binarization-processed and division- processed at every character, and it is judged by a character contour extracting part 12 which the direction of lines constituting the character is out of the four directions set at every 45 deg.. Further, based on the said decision result, a character generating part 13 generates the number of the lines for the respective four directions as the feature quantity, and while the reading conditions are being changed, the said series of the processings are executed for the characters in a K number respectively for M times. In addition, the feature quantities in a KXM numbers are averaged by a feature quantity generating part 13, and it is made into the dictionary data of the character.

Description

[Detailed Description of the Invention] [Summary] Regarding a method for creating a multi-font dictionary for recognizing multiple character fonts with a single dictionary, even a single dictionary can recognize many types of fonts at a high recognition rate. The purpose is to provide a method for creating a dictionary that can be recognized, and includes the steps of extracting the outline of a character made of binarized pixels, and dividing the area of one character into NXN (N is an integer) sections. step, and the direction to the next adjacent pixel when tracing the pixels constituting the contour line in each section sequentially is 4 from the reference direction.
A step of creating an N×N×4-dimensional feature amount for each character by indexing it in four directions at 5” intervals, and a step of creating a feature amount of N×N×4 dimensions for each character by indexing the feature amount corresponding to the multiple character fonts. creating an averaged feature amount;
(Constructed by making it a dictionary for connoisseurs.

[Industrial application field]

The present invention relates to a method for creating a multi-font dictionary that allows a single dictionary to recognize multiple character fonts.

In a type recognition device, a document is read through a scanner, the printed characters are identified and coded one by one with reference to a dictionary, and a °ζ document file is created. Furthermore, this makes it possible to configure a system that performs various operations such as display, or combining with a word processing function to process documents.

The dictionaries used in such type recognition devices can be
It is desired that it be able to perform identification with a high recognition rate.

[Conventional technology]

When recognizing printed characters (here, "printed characters" refers to the kanji code defined by JIS), standards are set by JIS for target character fonts. However, in reality, the fonts printed in newspapers, magazines, etc., and the fonts printed in word processors, printers, etc., are similar to the fonts specified by JIS, but the details differ depending on the manufacturer. .

Therefore, in conventional type recognition devices, "ζ" has its own dictionary corresponding to these various fonts.
This was used to recognize printed characters.

[Problem to be solved by the invention]

It is desirable for a print recognition device to be able to commonly handle documents consisting of various different fonts, but in this case, preparing a dictionary for each font would result in an enormous amount of data. Not practical.

Therefore, it is conceivable to use a single dictionary to recognize many types of phono I, but in this case, as mentioned above, each font has different fonts. If an attempt is made to recognize different character fonts using a correspondingly created dictionary, the recognition rate may deteriorate. Therefore, conventionally, there has been a problem that it is difficult to recognize many types of fonts using a single dictionary.

The present invention aims to solve the problems of the prior art, and provides a dictionary creation method that can recognize many types of fonts with a high recognition rate even with a single dictionary. The purpose is to

[Means to solve the problem]

The method for creating a multi-font dictionary according to the present invention, as shown in the principle structure in FIG. S3. S4 and
Step S5 of dividing the region of one character into N×N (N is an integer) sections, and determining the direction to the next adjacent pixel when sequentially tracing the pixels forming the outline in each section from the reference direction.
Step S6. Create a N×N×4-dimensional feature interval for each character by indexing corresponding to four directions at 5° intervals. S7, and step S8 of creating a feature set by averaging a plurality of features obtained corresponding to a plurality of character fonts, and forming a feature value group obtained by a character sample consisting of a predetermined number of different characters. This is a common dictionary for the plurality of character fonts.

[For production]

In the multi-font dictionary creation method of the present invention, as shown in FIG.
l), then convert this into a screen consisting of a set of black and white binary pixels corresponding to the shading of the image using an appropriate darkness value (
Step S2).

Next, such a text image is divided and cut out for each line of characters, and each line is further divided and cut out for each character constituting it (Ste Nobu S3).

Character contours are extracted for each character image separated in this way (step S4). Character contours can be extracted by a well-known method such as detecting boundaries between white and black pixels and black and white changes when scanning a screen in a certain direction.

Next, the region of one character is equally divided into NXN squares (step S5). Then, when the pixels on the contour line in each cell opening are sequentially traced from the top along the contour line, the direction to the next adjacent pixel is one of four directions at 45° intervals corresponding to the angle from the reference direction. Indexing in four directions is performed using indexes indicating whether or not (step S6).

Figure 2 shows indexes in four directions, with O being the right horizontal direction, and 1, 2.3 at every 45° counterclockwise.
Let us define the indices in four directions. Therefore, for example, when the direction to the next adjacent pixel is vertically downward, the index is 2.

In this way, when indexing is completed for all pixels on the contour line within a certain square opening, the numbers are added for each same number component, and the numbers are arranged in descending order of the angle from the reference direction to form a four-dimensional Let it be a feature quantity. For example, the feature amount is (2,
0,0,3), the number of exponents “O” is 2, the number of exponents “I” is 01, and the number of exponents “2” is 0111.
1. This indicates that the number of index "3" is three. Therefore, when there are no pixels within the square mouth, the feature amount is (
0°0.0.0).

By performing these operations for each square that makes up the character image and creating a 4-dimensional feature ■ for each square, a total of N x N x 4-dimensional features can be created for each character. (Step S7).

M
Do it twice. In other words, starting from the number of reads i=0, i>
Repeat this step by adding 1 to i until M is reached.
As a result, a feature group a is created.

Furthermore, in accordance with the method for creating the feature amount a, the same text in K types of fonts is read and the same processing is performed.

That is, by starting from the font number j=1 and repeating the process while incrementing j by 1 until j>K, a feature amount group b and a feature amount group C2- are obtained.

When the creation of K types of feature groups a, b, - is completed, the desired dictionary is obtained by averaging the feature amounts of all groups. That is, by adding the feature amounts of each group and dividing by °ζKXM, data obtained by averaging the feature amounts of each character can be obtained, and this is used as dictionary data for each character.

FIG. 3 shows an example of the configuration of a multi-font dictionary creation section corresponding to the basic configuration shown in FIG. That is, the document input is binarized and temporarily stored in the image memory 11. Next, the character contour extracting section 12 performs processing such as line cutting, single character cutting, and character contour extraction based on the data stored in the image memory 11. The feature amount generation unit 13 uses the data stored in the image memory II and the character contour data created by the character contour extraction unit 12 to divide N Processes such as creating N×N×4-dimensional feature quantities by indexing in four directions relative to the pixel direction are performed to create multiple feature quantity groups corresponding to each character font, and furthermore, the feature quantity of each group is Create dictionary data by averaging.

The created dictionary is output to and stored in an external storage device.

〔Example〕

FIG. 4 shows a processing flow of an embodiment of the present invention.

Further, FIG. 5 shows an example of a character font to be used as the target in this case.

In the embodiment shown in FIG. 4, a sentence (character sample) in which 3303 characters of the JIS Kanji code are arranged in code order is used as a document to be read from a scanner, and M=10. Character phono 1 used in this case
The type of is = 3, that is, as shown in Figure 5, FM
Font A (old JIS font) used in the 16β printer (Fujitsu type), Phono 1-B (new JIS font) used in the FMF printer (Fujitsu V), and Iwata Hoso Mincho font (for phototypesetting). There are three types of phono l-c (standard font), and therefore a total of K x M = 30 processes are performed and averaged to create a dictionary.

7. Processing follows the flow shown in FIG. will be carried out.

Each step in this case is indicated by the same number corresponding to each processing step shown in FIG. In other words, we binarize the read characters, cut out 1 line, cut out 5 characters, extract the contours of 8 characters, etc., and divide them into 8 x 8 grids to create 8 x 8 x 4 =
Create 256-dimensional features. Further, such a process is repeated 10 times while reading and changing the density etc. to obtain a feature group, and the same process is performed for three types of fonts, for a total of 30 times.

When all data is obtained, each special ff1rf group
Create dictionary data for each character by adding the numbers, dividing by 30, and averaging. Further, by performing such processing on each character constituting the character sample, a dictionary is created in which the character code is associated with the dictionary data of each character.

This dictionary associates each character that makes up a character sample with the averaged feature amount of each character,
When performing printed character recognition, the features of each character obtained by performing processing similar to that shown in Figure 4 such as character outline extraction and feature generation from the read characters are used in the dictionary described above. Print recognition can be performed by selecting the same or closest feature amount by comparing the two characters, and reading out the character code corresponding to this feature amount from the dictionary.

When each of the above-mentioned character samples was evaluated using this dictionary, a recognition rate of around 99% was obtained. This recognition rate sometimes exceeded the recognition rate when a dictionary was created and evaluated for each font.

FIG. 6 shows an example of the configuration of a printed character recognition device using the multi-font dictionary creation method of the present invention.
1 is a multi-font dictionary creation section shown in FIG. 3, and 14 is a scanner. Reference numeral 15 denotes a personal computer, which includes a central processing unit (CPU) 16 and a main storage device 17 that stores operating programs and calculation data for the CPU 16.

18 is an external storage device.

In FIG. 6, the scanner 14 reads a document into the multi-font dictionary creation unit 10, and the multi-font dictionary creation unit 10 extracts character outlines as described with reference to FIG.

Processing such as creation of feature quantities is performed. The personal computer 15 uses the multi-font dictionary creation unit 10 to create a special m1it for each character constituting the character sample.
This is taken out and recorded 12g in the external storage device 18.

In this way, the process is performed M times for a certain character font, and the same process is repeated for another character font of the same character sample. When KXM processes are completed, the personal computer 15 stores Dictionary data for each character is created by extracting all feature ■ group data from , adding them together, and averaging them using KXM. The created dictionary data is stored in an external storage device]
8 and stored. By performing such processing on each character constituting a character sample, a dictionary for that character sample is created.

〔Effect of the invention〕

As explained above, according to the present invention, a dictionary for recognizing printed characters is created by averaging feature values extracted from a plurality of character fonts, so a dictionary that emphasizes the common parts of each character font is created. Therefore, it becomes possible to recognize printed characters consisting of many types of characters phono I with a high recognition rate using a single dictionary.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the basic configuration, FIG. 2 is a diagram showing four-direction indexes, FIG. 3 is a diagram showing an example configuration of a multi-font dictionary creation section, and FIG. 4 is a process of an embodiment of the present invention. FIG. 5 is a diagram showing an example of a character phono I-, and FIG. 6 is a diagram showing an example of the configuration of a printed character recognition device to which the present invention is applied. 10.--Multi-font dictionary creation section 11--Image memory 12--Character ring 9μ extraction section 13--Feature box generation section 14--Scanner 15--PC 18--External storage device

Claims (1)

[Claims] Steps of extracting the outline of a character consisting of binarized pixels (S2, S3, S4), and a step of dividing one character area into N×N (N is an integer) sections (S5). N
Steps of creating ×N × 4-dimensional features (S6, S7
), and a step (S8) of creating a feature amount by averaging a plurality of feature amounts obtained corresponding to a plurality of character fonts. A method for creating a multi-font dictionary, characterized in that a common dictionary for the plurality of character fonts is used.