Image Retrieval by using Non-Subsampled Shearlet Transform and Krawtchouk Moment Invariants

Abstract

In this paper, we use non-subsampled shearlet transform (NSST) and Krawtchouk Moment Invariants (KMI) to realize image retrieval based on texture and shape features. Shearlet is a new sparse representation tool of multidimensional function, which provides a simple and efficient mathematical framework.We decompose the images by NSST. The directional subband coefficients are modeled by Generalized Gaussian Distribution (GGD). The distribution parameters are used to build texture feature vectors which are measured by Kullback–Leibler distance (KLD). Meanwhile, low-order KMI are employed to extract shape features which are measured by Euclidean distance (ED). Finally, the image retrieval is achieved based on weighted distance measurement. Experimental results show the proposed retrieval system can obtain the highest retrieval rate comparing with the methods based on DWT, Contourlet, NSCT and DT-CWT.