擬Zernikeモーメントを入力データとしたニューラルネットワークによる文字認識

概要

論文の詳細を見る
This paper describes a character recognition using pseudo-Zernike moments which are applied to a neural network. Pseudo-Zernike moments which are orthogonal complex moments are analogous to Zernike moments in process of derivation. In addition, it is possible to extract rotation inveriant features from those moments. Also pseudo-Zernike moments have strong recognition ability for the images degraded by deformation, noise, etc. The principal aim of this paper has the following two points : (1) the algorithm for the sake of increasing the recognition rate is shown and (2) the adequate range of the degree of the moments is obtained in the recognition experiments. As the results, the better outcomes by far than the ones in case of the traditional statistical classifier are obtained.
東海大学の論文