JPS6182282A - Pattern recognizing system - Google Patents

Abstract

PURPOSE:To recognize surely a wide range pattern without being limited to kinds of patterns by utilizing knowledge information reserved in advance to various input patterns so as to apply each pattern processing required for pattern recognizing processing. CONSTITUTION:A pattern processing circuit 201 applies a quantizing processing to an input pattern data f1 at a prescribed level and gives the result to a matching circuit 10 via a bus 26. The matching circuit 10 applies matching processing between the standard pattern stored in a dictionary 221 for quantization and the said input pattern, and the result of matching including the similarity is inputted to an identification circuit 12. The identification circuit 12 applies an identification processing to an input pattern based on a parameter stored in a parameter memory 231, the result of identification is inputted to a knowledge circuit 211 for quantization and the knowledge circuit 211 applies optimum quantization processing. The knowledge circuit 212-22n give the required information to pattern processing circuits 202-20n so as to execute optimum noise processing, detection and segmentation processing, normalizing processing and sampling processing.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a pattern recognition method for recognizing a plurality of types of patterns such as characters, sounds, figures, etc.

[Technical Prohibitions of the Invention and Problems thereof] In recent years, pattern recognition technology has been rapidly developing as a basic technology for data processing, and the scope of application of pattern recognition objects has been expanding. In general, pattern recognition technology involves feature optimization processing and various pattern processing (e.g. preprocessing, normalization, sampling, etc. processing) for the recognition target pattern.
is an important element.

By the way, the type of pattern to be recognized is the character.

High-level pattern recognition that is not limited to individual patterns such as graphics, sounds, etc. is one of the important goals of pattern recognition technology.

[Purpose of the invention] The object of the present invention is to have a simple configuration and to create 9 character shapes.

An object of the present invention is to provide a pattern recognition method that can configure an advanced pattern recognition system that can reliably recognize multiple types of patterns such as speech.

[Summary of the Invention] The present invention has a configuration in which knowledge information processing is applied to pattern recognition processing to optimize pattern processing necessary for various pattern recognition processing.

The type of pattern of the input pattern to be recognized is determined by the matching processing means. The input pattern is subjected to various pattern processing necessary for recognition processing.

At this time, knowledge information for optimizing various pattern processings is output from the knowledge information memory means according to the type of pattern determined as described above. The apparatus is configured such that recognition processing for the pattern obtained after the optimum pattern processing is executed.

With such a configuration, it is possible to reliably perform pattern processing according to each input pattern without being limited to the type of input pattern.

[Embodiment of the Invention] An embodiment of the present invention will be described below with reference to the drawings. 1st
The figure is a block diagram showing the basic configuration of a pattern recognition system according to an embodiment of the present invention. In Figure 1,
The matching circuit 10 performs matching processing (
For example, a composite similarity method process) is executed to determine the category of the input pattern. The standard patterns consist of various standard patterns corresponding to characters, figures, sounds, etc., and are stored in the standard pattern memory 11 in advance. The identification circuit 12 is
Identification processing is performed on the pattern of the category determined by the matching circuit 10 based on parameters necessary for identification. This parameter is stored in the parameter memory 13 in advance.

FIG. 2 shows a specific circuit of a pattern recognition system constructed based on the basic circuit as described above. In FIG. 2, a plurality of pattern processing circuits 20-1 to 20-n perform various pattern processing (quantization, noise, cutoff, normalization, sampling, etc.) necessary for pattern recognition processing on input patterns. This is a circuit that executes each process (processing). The knowledge circuits 21-1 to 21-n are each pattern processing circuit 20-1 to 21-n.
20-n is a circuit including a memory group in which knowledge information necessary for performing optimal pattern processing based on the identification result of the identification circuit 12 is stored in advance. Further, the dictionaries 22-1 to 22-n are memory groups in which various standard patterns necessary for the matching process of the matching circuit 10 are stored in advance. Parameter memory 23-1 to 23-
n stores in advance various parameters necessary for the identification process by the identification circuit 12. Further, 24 to 27 are buses, respectively.

In the pattern recognition system configured as described above, the operation according to one embodiment will be explained. First, the basic operation of pattern recognition processing will be explained with reference to FIG. character 1
Assuming that an input pattern belonging to either graphics, audio, etc. is output to the matching circuit 10, the 1° matching circuit 10 determines the category of the input pattern through matching processing with a standard pattern prepared in advance. That is, the matching circuit 10 is
Patterns of categories with high similarity values S are output to the identification circuit 12. At this time, the similarity value S is obtained from the inner product of the input pattern vector f and the standard pattern vector Φ. That is, the following formula (1) holds true.

82-(f, Φ)2...(1) Then, the identification circuit 12 selects an identification result pattern (vector λ) belonging to, for example, a character pattern from patterns with high similarity values based on parameters prepared in advance. It will be output.

Next, specific operations will be explained with reference to FIG. First, input pattern data (unknown pattern vector fx
) is output to the pattern processing circuit 20-1. Here, the pattern processing circuit 20-1 is a circuit that performs quantization processing at a predetermined level on an image signal that is input pattern data. The quantized input pattern is provided to matching circuit 10 via bus 26. Matching circuit 1
0, matching processing between the standard pattern stored in the quantization dictionary 22-1 and the input pattern is executed. The matching circuit 10 outputs a matching result (category of the input pattern) including the similarity value as described above to the identification circuit 12 . The identification circuit 12 performs identification processing on the input pattern based on the parameters stored in the quantization parameter memory 23-1.

When the identification result is output from the identification circuit 12, the quantum can knowledge circuit 21-1 determines the type of quantized pattern based on the identification result, and selects the optimal quantum according to the type of pattern. The information necessary for performing the quantization process (i.e., level values necessary for quantization, etc.) is sent to the pattern processing circuit 20-
Output to 1. As a result, the pattern processing circuit 20-1
performs optimal quantization processing on the input pattern for which the category has been determined, and then outputs the next pattern processing circuit 20-2.
For example, it is output to a noise processing circuit. The pattern processing circuit 20-2 executes noise processing and outputs the processing result to the matching circuit 10. In the matching circuit 10, matching processing is performed based on the standard pattern stored in the noise processing dictionary 22-2. When the identification circuit 12 outputs the identification result based on the matching result, the noise processing knowledge circuit 21-2 outputs information necessary for optimal noise processing to the pattern processing circuit 20-2. This results in
The pattern processing circuit 20-2 outputs the pattern after optimal noise processing to the next pattern processing circuit 20-3.

Similarly, each pattern processing circuit 20-3 to 20-
n allows each process, such as a cutoff process, normalization process, and sampling process, to be optimally executed. In this way, the final identification process is performed on the input pattern obtained after each optimal pattern process is performed, and a recognition result can be obtained.

[Effects of the Invention] As detailed above, according to the present invention, each pattern process necessary for pattern recognition processing can be executed for various input patterns using knowledge information prepared in advance. . With this, it is possible to realize a pattern recognition system that can reliably recognize a wide range of patterns without being limited to the types of patterns to be recognized, with a simple configuration.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the basic configuration of a pattern recognition system according to an embodiment of the present invention, and FIG. 2 is a block diagram showing the specific configuration of the same embodiment. 10... Matching circuit, 11... Standard pattern memory,
12... Identification circuit, 13... Parameter memory, 2
1-1 to 21-n...Knowledge circuit. −〇−

Claims (1)

[Claims] A matching processing means for determining the type of the input pattern by matching the input pattern to be recognized with a standard pattern prepared in advance; and an identification means for identifying the input pattern based on the determination result of the matching processing means. , a plurality of pattern processing means for performing various pattern processing necessary for recognition processing of the input pattern, and each of the pattern processing means executes optimal pattern processing for the input pattern based on the identification result of the identification means. and knowledge information storage means that stores in advance each piece of knowledge information for
A pattern recognition system characterized in that the processing results of each of the pattern processing means obtained based on the knowledge information of the knowledge information storage means are output to the matching processing means.