Sharpness for texture retrieval in multiscale domain

Sharpness is an important geometrical feature of 2D images which can be measured by statistic methods. FISH and FISH_bb are very efficient measurements of sharpness for a whole image. In this paper, we adopt and slightly modify FISH and FISH_bb as kinds of texture features, combined with traditional statistical features such as means and standard deviations, and apply these features for texture retrieval in multiscale domain. We evaluate these combined features in several different multiscale transforms such as discrete Wavelet transform, dual-tree complex Wavelet transform, Contourlet transform and pyramidal dual-tree directional filter banks. The texture retrieval scheme involving sharpness and multiscale achieves good performance with higher retrieval accuracy in numerical experiments, and these results show that sharpness is a kind of efficient texture feature for 2D images.