JPS5853390B2 - pattern recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a pattern recognition device, particularly for example, a pattern recognition device for identifying handwritten characters, which reduces the frequency of rejection of recognition target figures and prevents misreading. It is something.

As is conventionally known, as a pattern recognition device that identifies a figure to be recognized such as a handwritten character, each bit information of each standard pattern prepared in advance in a dictionary and each bit information of a feature pattern extracted from the figure to be recognized are used. There is a system in which, for example, when a match is achieved in all areas of focus, the figure to be recognized is recognized as a figure indicated by the standard pattern for which the match has been achieved.

In this case, consideration is given to creating a so-called learning effect by adding feature patterns that are not matched, that is, rejected, into the same dictionary as new standard patterns.

However, in this type of pattern recognition device, there is a high probability that a recognition target figure that should originally be correctly recognized will be misread.

That is, for example, if a standard pattern corresponding to the number "9" with a collapsed lower loop has been added to the dictionary, and a relatively correctly written number "9" is input as a recognition target figure, There is a possibility that the feature pattern extracted from the number "9" and the standard pattern corresponding to the number "9" may undesirably match and be mistakenly identified as the number "8".

The present invention aims to solve the above-mentioned problems, and aims to sufficiently reduce the frequency of rejection and the probability of misreading by providing the above-mentioned learning effect.

Therefore, the pattern recognition device of the present invention is equipped with a feature buffer in which feature patterns extracted from the recognition target figure are stored and a dictionary in which a plurality of standard patterns are prepared in advance, and a pattern recognition device for identifying the recognition target figure. A spare memory is provided in which the extracted feature pattern in the feature buffer is written as write data when a predetermined condition is satisfied, and the feature pattern in the feature buffer, each standard pattern in the dictionary, and the reserve memory are provided. It has a matching circuit that detects the match/mismatch with the feature pattern written in the memory.
The matching circuit detects a match or mismatch between the feature pattern in the feature buffer and the standard pattern in the dictionary, and then detects a match between the feature pattern in the feature buffer and the feature pattern written in the preliminary memory.・When a discrepancy is detected and a discrepancy is found in the detection.

The present invention is characterized in that the extracted feature pattern in the feature buffer is controlled to be written as a new feature pattern on the preliminary memory in association with separately input category information.

The present invention will be described below with reference to the drawings. The figure shows the configuration of an embodiment of a pattern recognition device according to the present invention.

In the figure, 1 is a feature buffer and 2 is a dictionary, such as a read-only memo! , l (ROM), and 3 is a spare memory, such as a random access memory (
4 represents a matching circuit, 5 represents a mask correction circuit, and 6 represents an operator operation section.

In the figure, a feature buffer 1 stores feature patterns extracted from features to be recognized.

On the other hand, in the dictionary 2, pairs of standard patterns and category information of the standard patterns are prepared in advance, for example.

Further, as will be described later, in the preliminary memory 3, a characteristic pattern of a figure that could not be recognized (hereinafter referred to as a preliminary standard pattern) and category information of the preliminary standard pattern are written in pairs.

The operation of the circuit shown in the figure will be explained below. The feature pattern stored in the feature buffer 1 is input to the matching circuit 4.

On the other hand, each standard pattern and each category information corresponding to each standard pattern are read out from the dictionary 2, and the matching circuit 4 reads out each standard pattern and each category information corresponding to each standard pattern.
is input.

The matching circuit 4 compares each bit information of the characteristic pattern with each bit information of each standard pattern.

[D] When each bit information of the characteristic pattern and each bit information of the standard pattern all match, a matching signal is output from the matching circuit 4, and the category 1. The figure to be recognized is identified based on the I'ltT information.

■ On the other hand, if the above feature pattern does not match any standard pattern in the dictionary 2, the preliminary standard pattern written in the preliminary memory 3 and the category information corresponding to the preliminary standard pattern are read out, and a match is made. In the circuit 4, each bit information of the characteristic pattern is compared with each bit information of the preliminary standard pattern.

Here, (II-1, 1) If the preliminary standard pattern that becomes the above-mentioned feature pattern matches, a matching signal is output from the matching circuit 4, and the category information of the preliminary standard pattern with which the above-mentioned match has been obtained is also used. A recognition target figure is identified.

(II-1') On the other hand, if the characteristic pattern does not match any of the preliminary standard patterns, (1) the matching circuit 4 outputs a reject signal.

(2) The reject signal is input to the mask correction circuit 5, and the mask correction circuit 5 outputs address information to the spare memory 3 and inverts the write control signal for the spare memory 3 to, for example, logic "1". to make the spare memory 3 into a writable state.

(3) Therefore, the preliminary memory 3 newly writes the characteristic pattern still stored in the characteristic buffer 1 as a preliminary standard pattern at the address indicated by the address information.

(4) At this time, the category information corresponding to the characteristic pattern is also written into the preliminary memory 3 in pairs with the preliminary standard pattern.

Here, the category information is supplied from the operator operation unit 6, and is given as a result of the operator recognizing the rejected figure to be recognized as described above using, for example, a video device.

(5) Then, the feature pattern in the feature buffer 1 is again compared with the standard pattern in the dictionary 2 and the preliminary standard pattern in the preliminary memory 3.

(6) As a result, the characteristic pattern now matches the preliminary standard pattern written in the preliminary memory 3 as described above.

(7) Therefore, a coincidence signal is output from the coincidence circuit 4, and the figure to be recognized is identified based on the category information of the preliminary standard pattern.

As described above, in the case of the present invention, the feature pattern of the figure to be recognized that could not be identified, that is, the preliminary standard pattern, is written in the preliminary memory 3, and each standard pattern in the dictionary 2 is referred to for the extracted feature pattern. Only when no match is obtained, the preliminary standard pattern in the preliminary memory 3 is automatically referred to.

Therefore, the probability of misreading the figure to be recognized can be sufficiently reduced, and the frequency of rejection can be sufficiently reduced.

It goes without saying that the pattern recognition device according to the present invention can achieve remarkable effects particularly when identifying handwritten characters written by the same person.

[Brief explanation of drawings]

The figure shows the configuration of an embodiment of the present invention, in which 1 is a feature buffer, 2 is a dictionary, 3 is a spare memory, 4 is a matching circuit, and 5 is a mask correction circuit.

Claims (1)

[Claims] 1. In a pattern recognition device for identifying the recognition target figure, which includes a feature buffer in which feature patterns extracted from the recognition target figure are stored and a dictionary in which a plurality of standard patterns are prepared in advance, the feature buffer is used as written data. A preliminary memory is provided in which the extracted feature patterns in the buffer are written when a predetermined condition is satisfied, and the feature patterns in the feature buffer, each standard pattern in the dictionary, and the feature patterns written in the preliminary memory are provided. A matching circuit is provided to detect a match/mismatch between the feature pattern in the feature buffer and the standard pattern in the dictionary, and the -number circuit detects a match/mismatch between the feature pattern in the feature buffer and the standard pattern in the dictionary. Matching/mismatching between the feature pattern and the feature pattern written in the above-mentioned spare memory is detected, and when a mismatch is found in the detection, a new feature pattern is written in the above-mentioned feature buffer as a new feature pattern on the above-mentioned spare memory. A pattern recognition device characterized in that the extracted feature pattern is controlled to be written in association with separately input category information.