JPS629958B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character recognition device, and particularly to an improvement for reducing the unreadable rate of a character recognition device using the degree of difference from a standard pattern as a criterion for determination.

When recognizing characters, the degree of dissimilarity (or, conversely, the degree of similarity) between multiple standard patterns and the input pattern is determined, and the value D ₁ with the smallest degree of dissimilarity (hereinafter referred to as the minimum value) is calculated. ) and category name
N ₁ , and the one with the minimum degree of difference among the category names different from this category name (because there is not necessarily one standard pattern for the same category) (referred to as the next smallest value D ₂ ) and its category name N ₂ , and if the minimum value D ₁ is less than a predetermined threshold T and the difference between the next smallest value D ₂ and the minimum value D ₁ is more than a predetermined threshold Td ₁ , determine the category name N ₁ . Category names were often used, and unless these conditions were met, they were often rendered unreadable. For example, Japanese Patent Publication No. 50-156322
The publication "Pattern Recognition Device" is one of them.
However, if you try to distinguish between the number zero 0 and the alphabetic letter O-O, or even the alphabetical characters O-O and D-D, which originally have similar shapes, for example, if the number 0 is entered, it will be recognized as the number 0. The difference between the degree of difference between the letter O and the letter O is very small, and as a result, it is often impossible to identify the letter O. Therefore, if the threshold value T _d1 is made small in order to reduce the rate of indiscernibility, a bad phenomenon occurs in that the number of misreadings increases. Adjusting this threshold value T _d1 is one of the most difficult things in character recognition device design.

As described above, an object of the present invention is to provide a character recognition device that reduces unreadability and does not increase misreading when similar characters occur.

In the present invention, even if the difference between the above-mentioned next minimum value D ₂ and minimum value D ₁ is less than the threshold value Td ₁ , these two values
If the category pair of category names N ₁ and N 2, which became _{D 1 and} D ₂ , is one of the pairs of category names previously determined to be similar, then the next minimum value D ₂ and the minimum The difference from the value D ₁ is the threshold Td ₂ (less than Td ₁ )
If it is above, it is good, and categories N ₁ and N ₂
In a category different from , the difference between the minimum difference D ₃ (hereinafter referred to as successively smaller values) and the minimum difference D ₁ is
Category name N ₁ if the condition of Td ₁ or more is satisfied
is output as the judgment result.

The present invention will be explained in detail below.

The figure is a block diagram showing an embodiment of the character recognition device of the present invention. One character on a document is quantized by the scanning quantization means 1 and the quantized input pattern is set in the difference detection means 2. The plurality of standard patterns stored in the standard pattern memory 3 and the input pattern are compared in the difference detection means 2, and the degree of difference with each standard pattern is calculated. This difference detection means is based on the above-mentioned Japanese Patent Application Publication No.
50-156322 "Pattern Recognition Device" etc., a method of calculating the degree of dissimilarity may be used, or
53-32658 "Character reading device" may be used to obtain similarity. In any case, the degree of dissimilarity and the degree of similarity are essentially the same, with the minimum degree of difference being the maximum degree of similarity, and therefore, in this application, they are referred to as the degree of dissimilarity. The dissimilarity ranking circuit 4 ranks the dissimilarities from the smallest to the largest, and sets only two sets of these dissimilarities and category names in candidate category registers 51, 52 and 53, 54. The third degree of difference is set in register 56. Furthermore, the category names of the standard patterns that have caused these three degrees of difference are made different.

The minimum value D ₁ and category name N ₁ are stored in registers 51 and 52, and the next minimum value D ₂ is stored in registers 53 and 54.
and the category name _N2 , and furthermore, the small value _D3 is set in the register 56 one after another. In this way, the ranking circuit 4, registers 51, 52, 53, 5
4 and 56, candidate categories are determined and output.

Category name set in register 51 N ₁
and the category name N ₂ set in register 53.
The category name checking circuit 61 checks whether the category name pairs match the category name pairs stored in the similar category pair table 62, and if a matching category name pair is detected, "1" is set in the register 63.

(Assume that this register 63 is reset to “0” before the test circuit operates) The similar category pair table includes (number 0 and alphabet O) (alphabet O and number 0) (8 and B) (B and Pairs of similar categories are stored as in 8). For example, if the input pattern is the number 0, register 5
In many cases, the number 0 is set to 1 and the alphabetic character O is set to register 53 as a candidate category.

In this case, since the category pair (number 0 and alphabetic letter O) exists in the similar category pair table, "1" is set in register 63. In addition, the contents of the table are memorized as (number 0 and alphabet O) (8 and B), and categories N ₁ and N ₂ are stored in the test circuit.
You can also replace it with and search again. register 52
The minimum value D ₁ set in the threshold value register 72 is compared with the threshold value T set in the threshold value register 72 in the comparator 71, and if D ₁ is less than or equal to T, the comparator 71
A signal “1” is output from.

The next minimum value D ₂ set in the register 54 and the minimum value D ₁ set in the register 52 are obtained by subtracting D ₁ from D ₂ in a subtracter 81 and obtaining the value.
D ₂ -D ₁ are input to comparators 83 and 84.

In the comparator 83, a threshold value register 87 is set. It is compared with the threshold Td ₂ and D ₂ − _{D 1} is
If Td ₂ or more, a signal “1” is output to the signal line 831. In comparator 84, threshold register 8
D _{2 −}
If D ₁ is greater than or equal to Td ₁ , a signal “1” is output to the signal line 841.

The successive small values D ₃ set in the register 56 and the minimum value D ₁ set in the register 52 are used in a subtracter 82 , where D ₁ is subtracted from D ₃ and the resulting value is set in a comparator 85 . In the comparator 85, D ₃ -D ₁ is compared with the threshold value Td ₁ set in the threshold value register 86, and if D ₃ -D ₁ is greater than or equal to Td ₁ , a signal "1" is outputted to the signal line 851.

When the test results of the similar category pair table are set in the register 63, these comparators 7
It is easy to obtain comparison results of 1, 83, 84, and 85. If the output signal line 631 of the register 63 and the outputs of the comparators 71, 83, 84 and 85 are all "1", the AND gate 9
The output of "1" becomes "1", and this signal "1" passes through the OR gate 93 and causes the category name N ₁ to pass through the gate 94. i.e. category name
When N ₁ and N ₂ are a similar category pair, the minimum value D ₁
is less than the threshold T, the difference between the next smallest value D ₂ and the minimum value D ₁ is more than the threshold Td ₂ , and the difference between the next smallest value D ₃ and the smallest value D ₁ is more than the threshold Td ₁ , then the category Name N ₁
is accepted as the judgment result.

If the signal “1” is not set in the register 63, that is, the signal line 631 becomes “0” and the signal line 63
2 becomes "1" and the comparison result of the comparators 71 and 83 is "1", the output of the AND gate 92 becomes "1", this signal "1" passes through the OR gate 93, and the category name _N1 is It is allowed to pass through gate 94. That is, if category names N ₁ and N ₂ are not in the similar category pair table, the minimum value D ₁ is less than or equal to the threshold T, and the difference between the next smallest value D ₂ and the minimum value D ₁ is greater than or equal to the threshold Td ₁ . In some cases, category N ₁ is determined as the determination result.

In this way, even when characters that are essentially similar, such as the number 0 and the alphabet O, are included in the reading target, it is possible to obtain a character recognition device that does not cause a decrease in the reading rate of these similar characters. can.

In the above explanation, we have only talked about alphanumeric characters, but it can also be applied to the recognition of kanji and katakana characters. .6 As described in "Machine Recognition of Handwritten Katakana Characters and Numbers - Phase Line Segment Method and Automatic Concept Formation-1", features are extracted from handwritten characters and judgments are made by determining the distance from the standard pattern. It can also be used in cases.

[Brief explanation of the drawing]

The figure is a block diagram showing an embodiment of the character recognition device of the present invention. 1 in the figure is a scanning quantization means, 2 is a difference detection means, 3 is a standard pattern memory, 4 is a ranking circuit, 51, 52, 53, 54
and 56 is a register, 61 is a category name search circuit, 62 is a similar category name table, and 8
1 and 82 are subtracters, 71, 83, 84, and 85 are comparators, 63, 86, 87, and 72 are registers, and 9
1,92 is and gate, 93 is or gate, 9
4 indicates an AND gate.

Claims (1)

[Claims] 1 Dissimilarity detection means for determining the degree of dissimilarity between an input pattern and a plurality of standard patterns with category names, the minimum degree of dissimilarity D ₁ among the dissimilarities, the category name N ₁ of the standard pattern, and the category Name N Minimum dissimilarity in categories other than ₁
D ₂ (hereinafter referred to as the next smallest value D ₂ ) and its standard pattern category name N ₂ , the category N ₁ and
Candidate category determining means for determining the minimum degree of dissimilarity D ₃ (hereinafter referred to as successively smaller value D ₃ ) in categories other than N ₂ , and the determined categories N ₁ and N ₂
is a similar category pair, and when the similar category pair checking means tests that the categories N ₁ and N ₂ are the similar category pair,
The minimum value D ₁ is less than or equal to a predetermined threshold T, and the difference between the successively smaller value D ₃ and the minimum value D ₁ is a predetermined threshold Td ₁
If the above three conditions are satisfied, and the difference between the next minimum value D ₂ and the minimum value D ₁ is greater than or equal to a predetermined threshold Td ₂ (<Td ₁ ), then the category name N ₁ is recognized as a judgment result and output, and if it is checked that categories N ₁ and N ₂ are not in a similar category pair, the minimum value D ₁ is set to a predetermined threshold T
In the following, if the difference between the next minimum value D ₂ and the minimum value D ₁ satisfies two conditions, that is, a predetermined threshold value Td ₁ or more, the category name N ₁ is recognized as a determination result and is output. A character recognition device characterized by: