Image Retrieval Method Based On Composite Texture Feature Representation

Abstract

Aiming at the insufficiency of multi-scale texture feature description ability of gray level co-occurrence matrix, based on the multi-direction and multi-scale feature of non-sub sampled shearlet transform, a new method of texture feature description is proposed. The composite texture feature consists of two aspects: on the one hand, the four statistic of the gray level co-occurrence matrix of the image is used as a local texture feature; on the other hand, the non-sub sampled shearlet transform is performed on the image to obtain different scales and directions. Based on the mean and variance of the energy of the sub-band coefficient image, the texture features of the image with multi-scale characteristics are formed. The two features are given different weights to form a composite texture feature, and the Euclidean distance is used as the similarity measure to realize image retrieval. This composite texture feature can not only represent the local texture feature of the image, but also express the multi-scale texture feature of the image. Experimental results show that the precision of image retrieval based on composite texture feature description method is fine.