JPS6152783A - Pattern recognizing system - Google Patents

Abstract

PURPOSE:To evade an erroneous recognition by extracting a feature amount of a pattern by only scanning in linear direction, carrying out an axis expansion matching based upon the dinamic programming modifying distorsion of the extracted pattern, and providing a pattern matching. CONSTITUTION:Peripheral feature amount in X, Y directions is calculated by accumulating the number of black dots every line in the directions of X and Y with respect to an input pattern of a binary image obtained by a pickup device and an input peripheral pattern 13 is stored. Correspondingly the result to each of 10-12 reference peripheral patterns corresponding to a plurality of reference patterns stored in a memory preliminarily stored, an axis expansion matching is carried out by a dinamic programming DP method to approach the distance between both the patterns as near as possible and obtain contrast patterns 13a- 13c. Absolute values of the respective differences between the contract patterns 13a-13c and the reference peripheral patterns 10-13 are calculated. An input pattern corresponds to the reference peripheral pattern 10 is recognized by one of obtained difference patterns 14-16.

Description

[Detailed description of the invention] [Technical field of invention] The present invention relates to a pattern recognition method.

[Technical background of the invention and its problems]

・When recognizing patterns such as characters and various figures by optically or electrically processing, the two-dimensional signal obtained by imaging the pattern is stored in memory and compared with a reference signal previously stored in another memory for discrimination. things are being done. Conventionally, when processing an imaged two-dimensional signal, two-dimensional array data is used as a target, and a portion having characteristics within the data is cut out and discriminated by matching it with a corresponding portion within a reference signal. However, such a method requires calculating the position (address) of a feature on a surface, which is a factor that impairs the speeding up of pattern recognition. Also,
If pattern matching is performed without changing some kind of correction based on the reference pattern to the input pattern, the problem arises that relative deviations and partial deformations cannot be ignored, leading to erroneous recognition. . In recent years, dynamic programming has been used as a method to deal with this misalignment problem.
matching (hereinafter abbreviated as DP matching) is used.

This Dr matching is mainly applied to the field of speech recognition, but it has also been attempted for character recognition. In this character recognition, since the overall matching is performed in row 9, there is a problem in that processing calculations for tracking the overall shape are complicated. Indeed, even if DP matching is applied to the above conventional method of cutting out characteristic parts within two-dimensional data, the problem of the burden of address calculation remains.

[Purpose of the invention]

An object of the present invention is to provide not only highly reliable pattern recognition but also a recognition method that significantly simplifies the recognition process.

[Summary of the invention]

It is configured to extract features in one-dimensional direction from the image signal (hereinafter referred to as peripheral features), and to perform axial expansion/contraction between the reference pattern and the input pattern with respect to a small amount of peripheral objects. be.

[Embodiments of the invention]

First, with reference to FIG. 1, a description will be given of the processing of a trace amount of peripheral objects by obtaining the peripheral distribution of a certain pattern and using it as a feature of that pattern. That is, the mountain is a pattern in which a binarized image obtained by an imaging device such as an ITV is stored in a memory.

For example, by summing up the number of black dots for each row in the X direction with respect to this pattern, a peripheral pattern in the X direction indicated by reference numeral (2) is obtained. This peripheral pattern (2) is assumed to be a trace amount of peripheral matter. If the same process is performed in the Y direction, a peripheral pattern (3) in the Y direction can be obtained.

An embodiment of the present invention will be described based on the processing of a small amount of peripheral objects as described above.

FIG. 2 shows a reference peripheral pattern α stored in a storage memory as a trace amount of peripheral objects for a plurality of reference patterns.
Q. (111, OA.The above reference pattern is one that has been standardized based on dimensional changes or graphical changes of a certain specific pattern, or one that has been selected from completely different types.) However, in this embodiment, the latter pattern is used.

Next, a pattern to be recognized is input, its surrounding feature amount is calculated, and the input surrounding pattern (to) is stored. This input peripheral pattern α→ is used as a reference peripheral pattern aO, αV.

The distance between both patterns is made as close as possible by performing cy-axis expansion and contraction using the DP thousand method corresponding to each αJ. As shown in Figure 3, the above-mentioned axis expansion and contraction results in a contrasting turn (xaa
), (13b), (13C) are obtained.

After the above processing, a comparison pattern (13a) is created.

(13b), (13c) and reference peripheral pattern αO9α
, and (6), that is, αG − (xa
a), (il)-(13b).

(b) Calculate the absolute value of - (13c). As a result, a difference putter with a weight of 70, (c) and α0 as shown in Fig. 4 is obtained, and among these, the difference pattern α distortion is smaller than the others;
Therefore, it is recognized that the input pattern corresponds to the original reference pattern of the reference peripheral pattern αQ.

Note that the scanning direction for calculating the amount of peripheral objects is not limited to the X or Y directions described above, but may also be obtained by scanning in an oblique direction or in a specific direction from the center of the pattern.

〔Effect of the invention〕

As mentioned above, we can extract the feature amount of the turn by scanning only in one-dimensional direction, and also perform axis expansion/contraction based on DP to correct the distortion of the extracted pattern, and perform pattern matching with t. This reduces the number of misidentifications in pattern recognition and makes it easier to perform pattern recognition.

[Brief explanation of drawings]

1 to 5 are diagrams showing various patterns for explaining one embodiment of the present invention. (10, α, ■...Reference peripheral pattern αa・
・・・・・・・・・・・・・・・・・・Input peripheral pattern agent Patent attorney Noriyuki Chika (and 1 other person)

Claims (1)

[Claims] In a pattern recognition method for measuring the degree of similarity between an input pattern obtained by an imaging device and a specific reference pattern among one or more reference patterns, the one or more reference patterns are scanned in at least one dimension. a step of converting the input pattern into a predetermined reference peripheral pattern by counting the number of characteristic pixels at intervals of , and storing it; and a step of converting the input pattern into an input peripheral pattern using the same method once the reference peripheral pattern is obtained. , comprising the steps of converting this input peripheral pattern into a comparison pattern by performing expansion/contraction axis alignment corresponding to each of the reference peripheral patterns, and counting the difference between these comparison patterns and the reference peripheral pattern,
A pattern recognition method characterized by recognizing similarity based on the magnitude of the difference.