等分散特徴パラメータを用いた文字識別法

概要

論文の詳細を見る
The most important problem in the field of pattern recognition is to find the system by which pattern dimensionality can be reduced without loss of spacial topological information on the pattern. We are concerned with the feature extraction problem. However, in this paper, a useful pre-processing technique for the reduction of pattern dimensionality is proposed; the proper linear space which is effevtive for the clustering of the objects is offered as a post-processing after the feature extraction. This pro-processing technique is the sampling technique which gathers mesh point values near the sampled mesh with Gaussian weights. The new space is equi-variance space for the enable of standard characters. The system mentioned here is very usful for the first steps of recognition systems. The proposed method has applied to pattern clustering of Japanese Katakana as an illustration; the result of a computer simulation are given here.
一般社団法人情報処理学会の論文
1971-01-15