Progressive rotation-invariant texture retrieval based on inter-scale dependency model
基于尺度相关性的旋转不变纹理图像渐进式检索算法

During the traditional wavelet rotation-invariant texture retrieval algorithms, the extracted directional information is limited and the inter-scale dependency between the coefficients is ignored, which affects the efficiency of retrieval. In this paper, the authors propose a novel progressive rotation-invariant texture retrieval algorithm based on inter-scale dependency. Firstly, Log-polar transform and Non-subsample Contourlet transform (NSCT) are combined to acquire rotation-invariant multi-scale and multi-orientation coefficients, then generalized Gaussian distribution (GGD) model is used to extract the global structure information from low-pass coefficients which can be employed further as coarse retrieval features. Afterwards, the Non-Gaussian Bivariate Model is employed to model NSCT coefficients inter-scale dependency, which can be used as fine progressive retrieval foundations. Finally,the performance of the algorithm proposed is illustrated by experiments based on Brodatz standard texture database. Compared to inner-scale model GGD based on wavelet coefficients retrieval algorithm, our method provides better efficiency and accuracy, which is proved to be an efficient rotation-invariant texture retrieval means.