JPH04153788A - Character recognizer - Google Patents

Abstract

PURPOSE:To reduce the memory capacity of a dictionary and to increase the character collating speed by selecting a specific feature parameter capable of identifying the minor dictionaries and the characters of these dictionaries as a feature parameter for collation of characters of each minor dictionary. CONSTITUTION:All collation subject characters (a) stored in a dictionary and the dictionary is sorted into plural minor dictionaries (b) for each prescribed feature pattern which applies the proper features to the characters (a) respectively. At the same time, a specific feature parameter capable of identifying the dictionaries (b) and the characters of these dictionaries is selected as a feature parameter used for collation and identification of characters of each dictionary (b). Thus it is possible to decrease the number feature parameters for collation and identification characters and also to increase the character collating speed without deteriorating the reliability for recognition of characters.

Description

[Detailed Description of the Invention] [Summary] To reduce the memory capacity of a dictionary and increase the matching speed without reducing the reliability of character recognition, regarding a character recognition device using pattern matching. In a character recognition device that recognizes input characters by extracting the characteristic parameters of input characters and comparing them with a dictionary, the character recognition device uses a dictionary to extract unique characteristics of all characters to be matched stored in the dictionary. In addition to classifying into multiple small dictionaries for each given feature pattern, specific features are used to identify small dictionaries and characters within small dictionaries as feature parameters for matching for character recognition in each small dictionary. Configure by selectively adopting parameters.

[Industrial application field]

The present invention relates to a character recognition device using pattern matching.

[Conventional technology]

As shown in Figure 1, a conventional character recognition device uses feature parameters extracted from input characters (Figure (a)) and a dictionary (Figure (b)).
), and the character with the matching feature pattern or the character with the closest feature pattern was recognized as the input character.

[Problems to be Solved by the Invention] In a dictionary used in a character recognition device, as the number of feature parameters for comparison increases, the number of determination factors increases, and the reliability of character recognition improves. However, increasing the number of feature parameters increases the memory capacity of the dictionary and increases the time required for matching. For this reason, it is necessary to reduce the memory capacity as much as possible, but if the number of feature parameters is simply reduced, there is a problem in that the reliability of character recognition decreases accordingly.

The present invention has been made under the above circumstances, and provides a character recognition device capable of reducing the memory capacity of a dictionary without reducing the reliability of character recognition, and increasing the matching speed. It is.

[Means to solve the problem]

The present invention provides a character recognition device that recognizes input characters by extracting feature parameters of input characters and comparing them with a dictionary, as shown in FIG. All the matching target characters (Figure (a)) are classified into multiple small dictionaries (Figure (b)) for each predetermined feature pattern that gives unique characteristics, and the characters used for matching for the character U in each small dictionary are Specific feature parameters that can identify small dictionaries and characters within small dictionaries are selected and adopted as feature parameters.

[For production]

By using a dictionary with the above structure, the number of feature parameters for comparison for character recognition can be reduced without reducing the reliability of character recognition. Therefore, the memory capacity of the dictionary is reduced and the matching speed is increased.

[Example] An example of dictionary creation in the character recognition device of the present invention will be specifically described.

In order to create a dictionary to be used in the character recognition device of the present invention, as a preliminary step, we collect as much learning data as possible for each of all characters to be matched to be stored in the dictionary, and as many (characteristic parameters as possible) for each character. Find the probability of occurrence.

FIG. 4 shows an example of the probability of occurrence of feature parameters.

Once the occurrence probabilities of the feature parameters as shown in Figure 4 are obtained for all matching target characters, based on this, the feature parameters with an occurrence probability of 50% or more are coded "1", and the occurrence probability is 50%.
By converting each feature parameter that is less than % to the code "0", a dictionary of feature patterns unique to each character consisting of "1" and "O" as shown in Figure 7 (b) for all characters to be matched can be created. create.

Once a dictionary for all matching target characters has been obtained as described above, the first step in creating a dictionary of the present invention is to first divide all matching target characters constituting the dictionary into a plurality of characters for each unique feature pattern. Classify into small dictionaries. To illustrate the classification method into small dictionaries using the dictionary shown in Figure 2 (a) (the same dictionary as in Figure 7 (a)), first, we focus on specific feature parameters marked with an asterisk. , characters with the same feature pattern formed by the feature parameters marked with * are extracted as one group. In the case of the example in Figure 2(a),
The characters “A”, “Ma”, and “P” are extracted as the same feature pattern, so the characters “A”, “Ma”, and “P” are extracted as 1.
As a group, one small dictionary shown in FIG. 2(b) is created. In the same way, while changing the position of the official seal so that the extracted characters do not overlap,
Characters with the same feature pattern are successively extracted as one group, and each group is classified as a small dictionary consisting of its own unique feature pattern.

In this way, all characters to be matched are grouped according to unique characteristic patterns and classified into a plurality of small dictionaries.

Next, as the second step of creating a dictionary according to the present invention, the number of feature parameters for comparison for character recognition is reduced for each of the classified small dictionaries.

That is, as shown in FIG. 3(a), the character "A" classified as one small dictionary in the first stage processing is
To explain using "Ma" and "P" as an example, first of all,
Excluding the feature parameters marked * that were used for the rate dictionary classification in the first step, * is applied as shown in Figure 3(b).
Only the unmarked feature parameters are left as feature parameters for matching for character recognition. The reason for excluding the feature parameters marked with an asterisk is that the feature pattern composed of the feature parameters marked with an asterisk is based on the characters ``a'', ``ma'', and ``p'' included in the small dictionary. This is because they form the same characteristic pattern, and there is no discernment for character recognition for each character in the dictionary.

Next, as shown in FIG. 3(C), from among the feature parameters of FIG. A small number of specific feature parameters that can identify each character in the own dictionary are selected, and the selected feature parameters are employed as feature parameters for comparison for character recognition. As a method for reducing the feature parameters, as described in detail below, the multiplicative product M of the feature stability S in each small dictionary and the feature difference degree between each small dictionary, that is, M = SXD (1 ) M: Multiplicative product S: Feature stability within the small dictionary D = Calculate feature difference between small and small books, and calculate a predetermined number of features within a predetermined threshold in descending order of multiplicative product M for each small dictionary. Just select the parameters.

(1) Feature stability S in the small dictionary The feature stability S in the small dictionary is a measure of the frequency of occurrence of each feature parameter in the small dictionary. For example, if we assume that the probability of occurrence of each feature parameter in a small dictionary is as shown in Figure 4, the closer the probability of occurrence is to 50%, the more uncertain the occurrence rate of the feature parameter is. It can be said that the closer the probability is to 0% or 100%, the more stable the appearance rate of the feature parameter is. Therefore, the feature stability S in the small dictionary is given by the difference between the probability of occurrence of the feature parameter and 50%, that is, S=(probability of occurrence of the feature parameter - 50).

For example, when calculating the feature stability S for the first feature parameter in FIG. 4, the probability of occurrence of the first feature parameter is 90%, so S-(90-5
0)=40, and it can be seen that this is a feature parameter with relatively high feature stability, that is, a feature parameter with high discrimination power for characters in the dictionary.

(2) Feature difference between small dictionaries The feature difference between small dictionaries is a measure of the degree of difference in the frequency of occurrence of feature parameters between small dictionaries.

Using the three small dictionaries A, B, and C shown in FIG. 5, the degree of feature difference between the small dictionaries will be specifically explained. Furthermore, the fifth
The occurrence probability value in the figure is the average value of the occurrence probability of each feature parameter.

In FIG. 5, for example, when comparing the first feature parameter between each small dictionary, the probability of occurrence in small dictionary B is 3%, which is close to 0%.

On the other hand, the other two small dictionaries A and C have an occurrence probability of 85% and 7.
8%, which is close to 100%. This means that the first feature parameter is a very effective feature parameter for distinguishing between the small dictionary B and the other small dictionaries A and C. Also, when comparing the ninth feature parameter between each small dictionary, the occurrence probability for each small dictionary is 7
5%, 88%, and 82%, all of which show similar values, and this ninth feature parameter is the same as that of the small dictionary A.
, B, and C. It can be seen that this is a feature parameter that has almost no meaning in distinguishing between B and C. Therefore, the degree of feature difference between small dictionaries is given by the average value of the difference in occurrence probability between corresponding feature parameters between each small dictionary.

For example, if we focus on the small dictionary B in FIG. 5 and calculate the degree of feature difference between the first feature parameter of the small dictionary B and the other dictionaries A and C, we will find that the small dictionary B and the small dictionary A, C
The difference value of the first feature parameter between is 82% and 75
%, and feature dissimilarity D= (82+75)/2=7L
It will be 5%.

Therefore, this first feature parameter is the small dictionary B
It can be seen that this feature parameter has an excellent ability to distinguish between the other small dictionaries A and C.

Using the feature stability S and the feature difference degree obtained as described above, the multiplicative product M=SXD is calculated using the above equation (1). The magnitude of the value of the obtained multiplicative product M indicates that the feature parameter has excellent discriminative power both for identifying small dictionaries and for identifying each character within the small dictionaries.

Therefore, for example, as shown in FIG. 3(C), a predetermined number of characteristic parameters (1, 7, 11 ，
156.401st feature parameter) is selected as the feature parameter for matching of the corresponding small dictionary. For other small dictionaries, a predetermined number of feature parameters optimal for matching in each small dictionary are selected in the same way. In the dictionary of the present invention obtained as described above, the number of feature parameters for comparison is reduced without reducing the reliability of character recognition, and the memory capacity is reduced accordingly.

FIG. 6 shows an embodiment of the character recognition device of the present invention constructed using the dictionary obtained as described above.

Reference numeral 1 denotes a dictionary memory in which the dictionary of the present invention described above is stored. 2 is an image scanner for reading input characters as digital character data, 3 is an image data storage memory for storing character data read by the image scanner 2, and 4 is for collating input character data. 5 is a dictionary matching recognition process for recognizing input characters by pattern matching the frame machine parameters extracted by the feature extracting unit 4 and the dictionary in the dictionary memory 1. 6 is a shared memory for temporarily storing various data during processing, and 7 is a microcomputer that centrally controls the processing operations of the entire character recognition device.

Input characters read from the image scanner 2 are stored as digital data in an image data storage memory 3, and a feature extraction section 4 extracts required feature parameters from the characters.

These feature parameters are stored in the shared memory 6, and pattern-matched with the dictionary in the dictionary memory 1 by the dictionary matching recognition processing unit 5, and the character whose feature pattern matches or the character whose feature pattern is closest is selected as the input character. In the case of the present invention, the dictionary stored in the dictionary memory is classified into a plurality of small dictionaries giving unique characteristics, as described above, and the number of feature parameters required for matching is as described above. Efforts have been made to minimize this as much as possible. Therefore, the matching time is shortened by that much, and character recognition can be performed at high speed.

[Effects of the Invention] According to the present invention, all the matching target characters stored in the dictionary are classified into small dictionaries, and the feature parameters for matching for character recognition are reduced, so that the accuracy of character recognition can be reduced. It is possible to reduce the memory capacity of the dictionary without reducing the memory capacity, and to increase the speed of matching.

[Brief explanation of the drawing]

Fig. 1 is a diagram showing the principle structure of the dictionary used in the present invention, Fig. 2 is a diagram showing an example of classification into small dictionaries, Fig. 3 is an explanatory diagram of the classification method into small dictionaries, and Fig. 4 is a diagram showing characteristics. A diagram showing an example of the probability of occurrence of a parameter, FIG. 5 is an explanatory diagram of the degree of feature difference between small dictionaries, FIG. 6 is a plot diagram of an embodiment of the character recognition device of the present invention, and FIG. 7 is a diagram of conventional character recognition. FIG. 2 is a diagram showing the dictionary structure and character matching principle in the device.

Claims (1)

[Scope of Claim] A character recognition device that recognizes input characters by extracting feature parameters of input characters and comparing them with a dictionary, wherein the dictionary is a character recognition device that identifies all matching target characters stored in the dictionary. In addition to classifying into multiple small dictionaries according to predetermined feature patterns that give characteristics, it is possible to identify small dictionaries and characters within small dictionaries as feature parameters for matching for character recognition in each small dictionary. A character recognition device characterized by selectively adopting specific feature parameters.