Co-occurrence Features of Multi-scale Directional Filter Bank for Texture Charcterization

In this paper, we propose to use co-occurrence components. Spatial correlation of wavelet coefficients due features computed from multi-scale directional filter bank to the structure of textures can be captured by the co(MDFB) for texture characterization. As the filter band occurrence features. In this paper, we propose to use cocoefficients are localized frequency components, features from occurrence features calculated from MDFB for texture co-occurrence matrices of filter bands can characterize description. The components of MDFB are also localized but structures of textures by describing correlation among with higher angular frequency resolution than wavelet. We coefficients. Our experiments show that the co-occurrence describe and explain the co-occurrence matrices computed features outperform energy features considerably in texture from MDFB subbands for feature extraction. Experimental retrieval. In particular, they significantly improve the retrieval results in texture retrieval are provided to show the rate for textures with weak directionality and periodicity while effectiveness of the proposed features for texture still maintain a high retrieval rate for regular textures as the characterization. energy features. This paper is organized as follows. In Section II, the MDFB adopted is introduced. Then, computation of