Chinese Journal of Computers   Full Text
Title	The Texture Image Retrieval Algorithm Combined Statistical Features of the Circular Symmetric Contourlet with Local Binary Pattern
Authors	LIAN Qiu-Sheng LI Qin KONG Ling-Fu
Address	(Institute of Information Science and Technology, Yanshan University, Qinhuangdao, Heibei 066004)
Year	2007
Issue	No.12(2198—2204)
Abstract &   Background	Abstract The circular symmetric contourlet transform(CSCT) overcomes the aliasing phenomenon of contourlet transform, and it has better direction selectivity than the contourlet transform. However, since the down-sampling operation in CSCT, it lacks the translation-invariant property which is essential for texture analysis. In this paper, the translation-invariant CSCT(TICSCT) is constructed by replacing the directional filter bank(DFB) in CSCT with undecimated directional filter banks(UDFB). The marginal distribution of TICSCT coefficients is modeled by generalized Gaussian density, and the similarity of two images is measured by Kullback-Leibler distance(KLD). Since visually distinct patterns may have matching subband statistics, KLD isn′t the complete texture descriptor. In this paper texture can be retrieved by combining KLD with local binary pattern (LBP) which is its complementary features to some degree. The experiments illustrate that the average retrieval rate of this method is 2.21%，4.02% and 0.89% higher than KLD, LBP and Gabor-Weighted-Euclid algorithm respectively. keywords contourlet transform; image retrieval; translation-invariant; Kullback-Leibler distance; local binary pattern background Content-based image retrieval is a hot research domain in information science. As one of the visual characters of images, texture is used in image retrieval broadly. So far the texture analysis methods could be classified three kinds: statistics method, structure method and model-based method. The sparse image representation plays an important role in content-based image retrieval. Since Gabor transform is similar to the receptive field of simple cell in mammal visual cortex, and can provided sparse image representation, the statistical features of Gabor transform was widely used in content-based image retrieval. In the past several years, the group concentrated on multiscale geometric analysis and its application under the support by the Hebei Province Education Bureau with the Title "Construction of multi-directional wavelet based on human visual system"(No.2004124). This project aims to design new multi-resolution sparse image representation which can capture the intrinsic geometrical structure such as edges and texture in natural images more effectively than standard separable wavelet transform. The group has gotten considerable achievement in this domain. They have constructed circular symmetric contourlet transform which has similar frequency partition with cortex transform. In this paper, The translation-invariant circular symmetric contourlet transform (TICSCT) which has similar frequency partition with CSCT is proposed. Since visually distinct patterns may have matching subband statistics, KLD isn′t complete texture descriptor. The authors combine KLD with local binary pattern (LBP) which is extracted in special domain. Better retrieval performance can be achieved when the combined features are applied in image retrieval. The image retrieval experiment shows its potential.