Contourlet versus Gabor Transform for Texture Feature Extraction and Image Retrieval

Abstract

Significant feature extraction for texture retrieval can be perfectly achieved using multiscale image decompositions, such as contourlet and Gabor representations. In this paper we compare the efficiency of contourlet decomposition variants and Gabor transform in terms of texture search and retrieval rates. Two distinct approaches, namely energy computation and generalized Gaussian distribution modeling are applied on multiscale image subbands for texture feature extraction and similarity measurement. Content-based texture retrieval experiments conducted on Vistex database, using Gabor Pyramid, contourlets and their redundant counterparts, confirm that the redundant contourlet transform (RCT) competitively improves the retrieval rate and achieves discriminant features.