JPH05233881A - Recognizing character table preparing method for character recognizing device - Google Patents

Abstract

PURPOSE:To prepare a high-reliability recognizing character table with massive character patterns concerning the recognizing character table preparing method for a character recognizing device. CONSTITUTION:The recognizing character table 12 is provided with a compared result storage area number applied character pattern storing means 6 equipped with a recognizing character candidate for each table number and a character pattern corresponding to the number of a compared result storage area for storing the compared result obtained by comparing the character pattern with a character recognizing dictionary, a compared result summing-up means 7 to input the compared result storage area number applied character pattern and to sum up the number of characters expressed by the character pattern for each compared result area number, and recognizing character table preparing means 9 to prepare a correspondence table 4' making correspondent the compared result storage area number and the table number of the recognizing character table and to prepare a recognizing character table 12 based on the summed-up result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of creating a recognized character table referred to in order to output a recognized character candidate in a character recognition device.

To recognize character patterns such as alphanumeric characters and katakana, the read character pattern is compared with a character recognition dictionary,
This is performed by referring to the recognition character table in which the number of the storage area in which the comparison result is stored and the recognition candidate character are associated with each other, and outputting the recognition candidate character.

The present invention provides a method for efficiently creating a highly reliable recognition character table using a large number of character patterns.

[0004]

2. Description of the Related Art FIG. 4 shows the structure of a character recognition apparatus to which the present invention is applied. In the figure, reference numeral 100 denotes a character pattern to be recognized, which is composed of print characters such as alphanumeric characters and katakana, and handwritten characters. Reference numeral 101 is an optical character reading device, and reference numeral 102 is a character recognition dictionary for comparing the read character pattern with a character pattern. A recognition candidate table 103 has recognition character candidates corresponding to the numbers of the storage areas of the comparison results of the character patterns and the character recognition dictionary. Reference numeral 104 is a character recognition means for converting the character pattern 100 into a character recognition dictionary 1
02, and the recognition candidate table 1 based on the comparison result.
03 is referred to obtain a recognized character candidate. 105
Is a recognition candidate output means for outputting the character candidates recognized by the character recognition means 104.

FIG. 5 shows a character recognition dictionary and a recognized character table to which the present invention is applied. In the figure, 110 is an example of a character recognition dictionary, which is a tree dictionary. The tree-type dictionary selectively traces each feature from rough features to detailed features of an image and uses the answer at the end as a comparison result. Reference numeral 111 is a correspondence table in which the number of the comparison result storage area at the branch end is associated with the table number of the recognition character table. Reference numeral 112 denotes a comparison result storage area, to which a number (terminal number) is given. 113 is a table number storage area,
The table number of the recognized character table corresponding to the comparison result storage area number is stored.

Reference numeral 114 denotes a recognition character table, which stores the recognition candidate characters in order by the table number corresponding to the terminal number of the comparison result of the character recognition dictionary 110.

Reference numeral 120 is a branch of the character recognition dictionary, for example, for determining whether or not there is a loop in the input image. If there is a loop, proceed to the next branch A and the loop If not, it proceeds to the next branch B.

In the configuration shown in the figure, when a character pattern representing a looped character (for example, the numeral "6") is input, there is a loop at branch 120 (YES).
And proceed to the next branch A. Characters without loops (eg,
If the character pattern representing the numeral "1") is input, since there is no loop, the process proceeds to the next branch B as (NO).

Such a comparison result is stored in the comparison result storage area 1
Suppose it fell to 12. Since the table number corresponding to the comparison result storage area 112 is 200, next, the table number 200 of the recognition character table 114 is referred to. The recognition character candidates stored therein are ranked first, “6”, second “mu”, and third “4”, and output as a recognition result.

FIG. 6 shows a conventional method of creating a recognized character table. Conventionally, in the recognition character table corresponding to the pattern comparison result by the tree type dictionary as described above, a worker manually analyzes the tree type dictionary and guesses the character pattern falling in the comparison result storage area at each end of the tree. The candidate character was decided by.

FIG. 6 shows a conventional method of manually creating a recognition character table. (1) The operator compares the branches of the tree dictionary with the characteristics of the character pattern. (2) Select the branch direction according to the characteristics of the character pattern.

(3) If the analysis has not been completed until the final branch, the processing from (1) is repeated. (4) When the analysis is completed up to the last branch and the comparison result storage area is obtained, it is assumed that the comparison result of the analyzed character pattern falls in the obtained comparison result storage area, and the analyzed character pattern is displayed in the recognized character table. Register characters as recognition character candidates.

[0013]

Even with a character recognition dictionary (tree dictionary) for alphanumeric characters, kana characters, etc., the number of storage areas for comparison results at the end is about tens of thousands. As described above, manually determining the storage location of the comparison result by following the branch requires a huge amount of time. Therefore, it was not possible to analyze a large number of character patterns, and the conventional character recognition table was not statistically supported.

An object of the present invention is to provide a method of creating a highly reliable recognition character table by statistically processing a large number of character patterns.

[0015]

According to the present invention, first, a character pattern in which a meaning character is known is compared with a character recognition dictionary, and the number of the comparison result storage area (the terminal number of the tree-type dictionary is assigned to each character pattern). ) Is added to prepare a character pattern to which the comparison result storage area number is added. Then, the number of characters corresponding to the character pattern is statistic for each comparison result storage area number, and the recognition character candidate of the recognition character table is determined based on the statistic result.

FIG. 1 shows the basic configuration of the present invention. In the figure, reference numeral 1 is a character pattern input means, which comprises a disk device or the like for storing a large number of character patterns whose characters (answers) to be expressed are known. A character recognition unit 2 refers to a character recognition dictionary on the basis of input character pattern data to obtain a comparison result. Reference numeral 3 is a character recognition dictionary reference means for referring to the character recognition dictionary for the input character pattern and obtaining a comparison result.

Reference numeral 4 is a character recognition dictionary, for example, a tree type dictionary, which has a comparison result storage area at the end. Reference numeral 4'is a correspondence table in which the number of the comparison result storage area and the table number of the recognition character table are compared.

Reference numeral 5 is a comparison result storage area number giving means for giving the number of the comparison result storage area stored as the comparison result with the character recognition dictionary to the compared character pattern. Reference numeral 6 is a comparison result storage area number-attached character pattern storage means for storing the comparison result storage area number in the attached character pattern. Reference numeral 7 denotes a comparison result totaling means, which inputs the character pattern of the comparison result storage area number-added character pattern storage means 6 and for each comparison result storage area number (terminal number), indicates the character represented by the stored character pattern. It is to count. A totaling result storage unit 8 stores the totaling results.

Reference numeral 9 is a recognition character table creating means,
Decide the recognition character candidate based on the totaled comparison result,
The recognition character table is created. Reference numeral 10 is a correspondence table creating unit that associates the comparison result storage area number with the table number of the recognition character table creating unit 11. Reference numeral 11 denotes a recognition character table creation unit, which determines recognition character candidates based on the aggregated comparison results and creates a recognition character table.

Reference numeral 12 is a recognition character table, which stores the recognition candidate characters in order by table number. Reference numeral 13 is a reject condition inputting means, which is a reject condition for not deciding a recognized character candidate when it is difficult to decide the first-ranked recognized character candidate because of a variation in the counting result in the comparison result storage area number. Is to be entered. Reference numeral 14 is a reject condition setting means, which holds the inputted reject condition.

[0021]

The operation of the basic configuration of FIG. 1 will be described. (1) Processing up to creation of the comparison result storage area number-added character pattern storage means 6.

Data (for example, print characters such as alphanumeric characters, kana characters, symbols, and handwritten characters) (for example, read by an optical reader and stored in a magnetic disk or the like) from the character pattern input means 1. Input data) to the character recognition means 2. Character recognition means 2
Refers to the character recognition dictionary for the character pattern input by the character recognition dictionary reference means 3 (in the case of a tree type dictionary, the character pattern data is input to the end of the tree). Then, the number of the comparison result storage area obtained for each character pattern (the end number of the tree in the case of a tree dictionary)
To get

The comparison result storage area number giving means 5 gives the comparison result storage area number referred to by the character recognition means 2 to the corresponding character pattern. Then, the comparison result storage area number-added character pattern storage means 6 stores each character pattern provided with the comparison result storage area number (end number of the tree dictionary) obtained by comparison with the character recognition dictionary. (If the character pattern input means 1 is a disk, the terminal number is given to the character pattern stored in the disk).

(2) Processing for creating recognition character table The comparison result totaling means 7 is for each comparison result storage area number (end number) for the character pattern stored in the comparison result storage area number-added character pattern storage means 6. Then, the characters and the number of characters represented by the character pattern are totaled.

The correspondence table creating section 10 associates the comparison result storage area number (terminal number) with the table number of the recognition table. The correspondence table 4'of the character recognition dictionary is created. The recognition character table creation unit 11 retrieves the collection data for each comparison result storage area number from the collection result storage unit 8 and registers the character represented by the character pattern as a recognition character in the table number created by the correspondence table creation unit 10. For example, end number 1 shown
As you can see in the tabulation result of, the answer letter "A" is 200
When there are 50 “P” s and 10 “Ts”, “A” is the first recognition candidate, “P” is the second recognition candidate, and “T” is the third recognition candidate. To register.

At this time, the recognized character table creating unit 11 refers to the reject condition and, for example, if the first character is less than 1/2 of the total number of characters of the corresponding terminal number, it is not registered as a recognized character candidate. If a reject condition has been set, the condition is referenced, and if the reject condition is satisfied, such as the end number 2 shown in the figure, it is determined that there is no recognition candidate in the column of the first recognition candidate of the corresponding table number. Register the reject.

According to the present invention, the recognized character candidates are determined by statistical processing from the large number of character patterns stored in the data file in the comparison result storage area number-added character pattern storage means 6. Therefore, a highly reliable recognition character table can be easily created.

Rejection conditions can also be set easily and recognition character tables for various conditions can be easily created by changing the rejection conditions.

[0029]

FIG. 2 shows an embodiment of a method for creating a comparison result storage area number-added character pattern storage means of the present invention.

In the figure, reference numeral 20 is a character pattern storage means for storing several million to 10 million kinds of alphanumeric characters, kana characters and symbols. Reference numeral 21 is an example of a character pattern, and 22 is an end number storage area for storing a comparison result storage area number (hereinafter, referred to as an end number) of a character pattern.

Reference numeral 23 is a character recognizing means, and 24 is a character pattern extracting means for extracting the character pattern of the character pattern storing means 20. Reference numeral 25 is a character recognition dictionary reference means.

Reference numeral 26 is a character recognition dictionary, which is a tree-type dictionary. Reference numeral 27 is a correspondence table, 28 is an end number storage area of the correspondence table, and 29 is a table number storage area of the correspondence table. Three
0 is a terminal number (recognition result storage area number) giving means.

Numeral 20 'is a terminal number (recognition character storage area number) -added character pattern storage means, which represents the terminal number written in the terminal number storage area of the character pattern in the character pattern storage means 20. 21 'is an example of the character pattern and is the same as the character pattern 21. Reference numeral 22 'is an end number storage area in which the end number is written.

The case of processing the character pattern 21 in the figure will be described. The character pattern extracting means 24 reads the character pattern 21. Character recognition dictionary reference means 2
5 inputs the read character pattern 21 into the character recognition dictionary 26. The character recognition dictionary 26 selects each branch according to the characteristics of the character pattern 21 and obtains a comparison result. The comparison result is stored in the terminal number storage area 28 (terminal number 500 in the figure).

The terminal number (recognition result storage area number) giving means 30 stores the terminal number 500 in the terminal number storage area 22 (22 ') of the character pattern 21 (21'). The above processing is performed for each character pattern, and the terminal number (comparison result storage area number) attached character pattern storage means 20 '
To create.

FIG. 3 shows an embodiment of creating a recognized character table according to the present invention. In the figure, 40 is a character pattern storing means for adding terminal numbers, 41 is a character pattern extracting means, 4
Reference numeral 2 is a comparison result totaling means, and 43 is a totaling result storage area.

Reference numeral 44 is a character recognition dictionary, which is a tree dictionary. 45 is a correspondence table, 46 is a comparison result storage area, 4
7 is a table number. Reference numeral 48 is a recognition character table creating unit, 49 is a correspondence table creating unit, and 50 is a recognition character table creating unit.

Reference numeral 54 is a recognition character table, 55 is a reject condition setting means, and 56 is an input means, which comprises a keyboard for inputting reject conditions. In the configuration shown in the figure, the character pattern extracting means 41 extracts the character pattern and the terminal number from the terminal number-added character pattern storing means 40 and inputs them to the comparison result totaling means 42.

The comparison result totaling means 42 totals the number of character patterns corresponding to the characters represented by the character patterns for each terminal number. In the illustrated example, the character "7" is displayed at the terminal number 100.
Indicates that there are 200 character patterns, the character pattern of the character "ku" is 50, and the character pattern of the character "wa" is 30. Similarly, the terminal number 101 has 100 character patterns of the character "2", 5 character patterns of the character "Yu",
The case where the character pattern of the character "K" is 5 is shown. The terminal number 102 indicates that there are 50 character patterns of the character “P”, 50 character patterns of the character “A”, and 5 character patterns of the character “D”.

In the recognition character table creating means 48,
The correspondence table creation unit 49 uses the comparison result storage area 46 of the correspondence table 45.
Correspondence table 45
To create. Then, the totalized result is fetched by the comparison result totaling means 42, and the recognized character table creation unit 50 refers to the reject condition (described later) and creates the recognized character table 54 based on the comparison result.

When the number of first-ranked character patterns is not more than 1/2 of the total number of terminal numbers, the character is not recognized as a recognized character candidate, and 1/2 is input from the input means 56 as a reject condition. Then, the reject condition is held in the reject condition setting means 55. In that case, if the number of first-ranked characters is 1/2 or less of the total number of terminal numbers in the counting result, the recognized character table creation unit 50 causes the recognized character table 5
Write the reject in the first column of 4. If the number of first-ranked characters is 1/2 or more of the total number of terminal numbers,
The candidate characters are registered in the recognized character table in descending order of the number of characters and registered. For example, the candidate ranks are registered up to the eighth rank.

[0042]

According to the present invention, it is possible to automatically create a recognition table based on a large number of character patterns. Further, the determined recognized character candidates are highly reliable because they are obtained by statistically processing based on a large amount of data.

Further, since the reject condition can be variably set, it is possible to easily create many kinds of recognition character tables according to different conditions. In addition, by using only numbers as the data to be processed, it is possible to easily create a dedicated character recognition table for each type of character, such as creating a character recognition table for only numbers.

Further, in the case of the tree type dictionary, the recognition character table itself is the total result of the tree type dictionary, so that it can be used as a material for improving the tree type dictionary, and the dictionary can be easily improved.

[Brief description of drawings]

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

FIG. 2 is a diagram showing an embodiment of a method of creating a comparison result storage area number-added character pattern storage means of the present invention.

FIG. 3 is a diagram showing an example of creating a recognized character table of the present invention.

FIG. 4 is a diagram showing a character recognition device as a target of the present invention.

FIG. 5 is a diagram showing a character recognition dictionary and a recognized character table, which are objects of the present invention.

FIG. 6 is a diagram showing a conventional recognition character table creating method.

[Explanation of symbols]

 1: Character pattern input means 2: Character recognition means 3: Character recognition dictionary reference means 4: Character recognition dictionary 4 ': Correspondence table 5: Comparison result storage area number giving means 6: Comparison result storage area number giving character pattern storage Means 7: Comparison result totaling means 8: Totalization result storage section 9: Recognized character table preparation means 10: Correspondence table preparation section 11: Recognized character table preparation section 12: Recognized character table

Claims (3)

[Claims] 1. A method for creating a recognition character table in a character recognition device for comparing a character pattern with a character recognition dictionary (4) and referring to a recognition character table (12) based on the comparison result to output recognition candidate characters, The recognition character table (12) has a recognition character candidate for each table number, and a character pattern in which the number of the comparison result storage area that stores the comparison result obtained by comparing the character pattern with the character recognition dictionary (4) is associated. The comparison result storage area number-added character pattern storage means (6) and the comparison result storage area number-added character pattern are input, and the number of characters represented by the character pattern is totaled for each comparison result storage area number. Aggregating means (7),
A correspondence table (4 ') in which the comparison result storage area number and the table number of the recognition character table are associated with each other is created, and a recognition character table creating means (9) for creating a recognition character table based on the counting result is provided. A character recognition device characterized by totaling the number of characters represented by a character pattern for each terminal number based on a comparison result storage area number-added character pattern and creating a recognition character table (12) based on the totalized result. For creating a recognition character table in. 2. The recognition character according to claim 1, further comprising reject condition input means (13) for inputting a rejection condition for not setting a recognition candidate character according to the comparison result, and reject condition setting means (14). The table creation means (9)
The recognition character table in the character recognition device characterized in that the rejection condition is compared with the aggregation result of the comparison result aggregation means, and if the aggregation result does not satisfy the rejection condition, no recognition character is registered in the recognition character table (12). How to make. 3. The recognition character table in a character recognition device according to claim 1 or 2, wherein the character recognition dictionary (4) is a tree-type dictionary and the comparison result storage area corresponds to the end of the tree. How to make.