心理学的特徴量による一般図形の識別モデル

抄録

It is the most important problem in non-symbolic pattern recognition of general shapes to decide what kinds of features should be extracted. Since the recognition of ‘general shapes’ depends on the purpose the recognizer has in mind, the ‘general shapes’ cannot be dealt with in general, while symbolic patterns such as letters of alphabet connote their categories a priori, the features that declare each symbolic patterns are evident.
On the other hand, the human being can discriminate not only symbolic patterns but also ‘general shapes’, for example, the random shapes dealt with in this paper. The authors consider the discrimination of the human being as the basis of recognizing of ‘general shapes’.
In this paper, the authors propose psychological experiments to find out what feature properties are given to the eyes of human subjects when they discriminate patterns, i. e. 100 convex and concave random shapes. And the standard searching shapes (categories) are set from the response of the subjects.
A multilevel mechanical procedure for discriminating the standard searching shapes (categories) is proposed. In the procedure the feature properties selected through the psychological experiments are used. It is shown that random shapes can be discriminated by this mechanical procedure in close agreement with a human judgement.