摘要
ＮＯＶＥＬＭＥＴＨＯＤＦＯＲＴＥＸＴＵＲＥＤＩＳＣＲＩＭＩＮＡＴＩＯＮＵＳＩＮＧＷＡＶＥＬＥＴＳ＊ＬｉｕＪｉａｎｆｅｎｇ（刘建峰）ＱｉＦｅｉｈｕ（戚飞虎）（Ｄｅｐｔ．ｏｆＣｏｍｐｕｔｅｒＳｃｉｅｎｃｅ，ＳｈａｎｇｈａｉＪｉａｏｔｏｎｇＵｎｉｖ．）Ｊｏ...
A novel multiresolution approach to the discrimination of textures using wavelets is presented. The approach employs an overcomplete wavelet decomposition called wavelet frames, which gives the description of both translation invariance and stability. In order to adapt to the quasi periodic property of textures, we propose the application of tree structured wavelet packet analysis. For discriminating efficiency, we develop a progressive texture discriminating algorithm, in which the discrimination process terminates once a suitably chosen discrimination criterion is met. Experiments show that with a minimal number of wavelet frame decompositions and iterations, our proposed approach could achieve a 100% correct discrimination rate on all the 12 texture types tested. That outperforms many of the existing approaches in terms of accuracy and computational efficiency, and thus it appears to be attractive for real time application involving texture based video/image classification.
