∇²Ｇフィルターを用いた多重解像度特徴抽出

概要

論文の詳細を見る
Feature extraction is an essential part of pattern recognition. A substantial amount of work is being done to obtain some features which are robust for distortions of character. However, there is no feature extraction method which exceeds the ability of human visual system. In this paper, we propose a multiscale feature extraction using Gaussian- Laplacian,∇²Ｇ , filter for handwritten character recognition. ∇²Ｇ filter can extract coarse-to-fine features in parallel. Moreover,this filter has an analogy with the human visual system. Therefore, our method has the opportunity of the same ability as human. In this approach,we show that same parameters can be automatically fixed using Akaike lnformation Criteria, AIC. The effetciveness of our method is examined with some experiments of character recognition.
大島商船高等専門学校の論文