Shun-ichi Kamemaru

Character recognition by feature extraction using cross-correlation signals from a matched filter

A hybrid pattern recognition system is described. It is based on a matched spatial filter (MSF) synthesized not by conventional template-matching patterns but by feature-extracted reference patterns. Some features are extracted from target objects used to recognize unknown objects. The optimum reference patterns for the filter are selected according to computer simulations. Through more than 30 simulations, four best patterns are chosen for recognition of 10 kinds of digits by the MSF. Recognition is performed not by conventional autocorrelation peaks but by cross-correlation signals because the reference patterns are not wholly the same as the input. The proposed system shows feasibility for feature-extracted pattern recognition using the multiplexed MSF.

Multimatched filtering using a microlens array for an optical–neural pattern recognition system

In the optical-neural recognition system proposed for flexible parallel information processing, a planar microlens array is used to form simultaneously many identical Fourier transforms of an input pattern; from each transform the feature extraction of the input pattern is performed in parallel through optical correlations with memorized standard spatial filters. In this paper, it is experimentally shown that the multimatched filtering system, as the optical feature extracting part of the proposed system, can be composed of a planar array of graded index microlenses of 1.05-mm diameter and 2.6-mm focal length.

A parallel-processing optical-digital recognition system as a model of biological visual perception

Abstract An optical-digital pattern recognition system is proposed as a model of biological visual perception. The optical part of this system extracts in parallel the geometrical features of each divided part of a pattern by using multi-matched filters made by an integrated small lens array. The fundamental experiments on the miniaturized matched filter of reflection type with a small lens of 3.3 mm diameter is reported, which show the feasibility to achieve the optical part of the proposed system.

Optical pattern recognition with object-multiplexed reflection-type matched spatial filters

Abstract Simultaneous recognition of the presence and the location of several desired patterns is performed using object-multiplexed reflection-type matched spatial filters (MSFs). The processing system produces the output recognition signals overlapped by the impulse response of the MSF, which makes parallel recognition easier. Experimental results and discussions are given.

Universal pattern recognition by matched filters synthesized by primitive patterns and by the algorithm for uniquely selecting the optimum reference patterns

In most pattern recognition systems, many target patterns should be recognized at one time by fewer reference patterns when the system uses a matched spatial filter. In this paper, two approaches are described for reducing the number of reference patterns in matched spatial filtering called universal pattern recognition which means a recognition technique of various shapes of patterns not depending on a target shapes. One approach is based on very simple bar patterns for the reference object for a matched filter according to a concept that every pattern or symbol is synthesized by many line components with various directions. By also using correlation diagrams with such reference patterns, 26 English alphabets were fairly recognized. Another technique is based on the algorithm to find automatically the unique and proper reference patterns of the matched spatial filter for recognition of the desired target sets. By the algorithm, unique and minimum numbers of reference patterns are selected and 26 English alphabets and 10 digits were recognized.

Pattern recognition by primitive and subtracted patterns as a reference object for a matched spatial filter

In this paper, a new approach for uniquely selecting the optimum reference patterns of a matched spatial filter (MSF) is described. By the reference patterns we can synthesize the optimum MSF despite the target patterns are changed from letters to numerals, for example. In the experiment, two types of the reference patterns are used. One of the types is called primitive patterns which contain a vertical bar, horizontal bar and circle patterns. They are all orthogonal to each other in the view of image processing which means that three patterns are not similar shape at all. The other type patterns are given by subtraction of a few target patterns which the primitive reference patterns can not recognize. Using these reference patterns, an MSF was synthesized which could fairly recognize the 26 English alphabet letters.

Pattern recognition system by a matched spatial filter and neural networks for imperfect input characters

A hybrid pattern recognition system using a matched spatial filter and neural networks is described. In the system, a multiplexed matched spatial filter synthesized by feature-extracted reference patterns is used for optical processing and an algorithm of a neural network is used for digital processing. The concept of the neural is based on an associative memory. It enables the system to recognize unknown input patterns with slightly different shapes from the reference patterns, which we could not distinguish by a conventional matched spatial filter. Using the system, imperfect input letters such as given by poor printing are fairly recognized as well as the perfect ones are recognized.

Fabrication of a character recognition system by a multiplexed matched spatial filter by feature-extracted patterns

In a conventional pattern recognition system with a matched spatial filter (MSF), the system is based on a template matching which is the comparison of the test pattern with a number of stored patterns until an exact match is found. Therefore, sometimes, the system has not been able to distinguish some similar patterns. In this paper, we propose to apply a concept of feature extraction to matched spatial filtering to improve performance of the MSF for distinction of some closely similar patterns. The synthesis of the MSF with partial patterns of unknown input objects to be recognized enables filtering operation to extract desired features from some input patterns. Using this technique and a hybrid pattern recognition system with the MSF synthesized by nine feature extracted patterns, we could perfectly recognize a page of unknown 25 alphabets. The recognition result was given not by conventional peaks but by character symbols on a CRT display.

Matched spatial filtering by feature-extracted reference patterns using cross-correlated signals

In this paper, a hybrid pattern recognition system is described that is based on a matched spatial filter (MSF) synthesized by feature-extracted reference patterns. Some features are extracted from test objects by computer simulations to find the optimum reference patterns for the filter. Through more than 30 simulations, four best patterns are chosen for recognition of 10 kinds of digits by the MSF. In the system, recognition is performed not by conventional autocorrelation peaks but by cross correlation signals because reference patterns for the filter are not exactly the same as the input unknown. The proposed system shows good feasibility for feature extracted pattern recognition.