Research On Image Retrieval Based On Multi-resolution Co-occurrence Matrix Texture Feature

With the rapid development of Internet, communication technology and storage technology, a lot of pictures emerge. How to retrieve the required image efficiently and accurately from a large amount of images is an important issue in image analysis and application fields. Content-based image retrieval technologyis a very effective method for retrieving huge amounts of data.The texture is one of the key elements characteristic description of the image.Wavelet transform has good time-frequency analysis capabilities, It has been widely used in image retrieval. But the traditional wavelet transform has limited direction selectivity to capture edge features and cannot effectively express the geometric characteristics of the edge of the image. Nonsubsampled Shearlet Transform (NSST) exhibits highly directional sensitivity and shift invariance, even it can be sparse representation of the image. In contrast, NSST acquires the natural texture and edge information compared with the traditional wavelet and has higher computational efficiently compared withContourlet.This paper studies the relevant properties of the nonsubsampled shearlet transform and its application in image retrieval,First it analyzes the statistical characteristics of NSST coefficient,and we can get the gaussian mixture model by using mixed gaussian function ondetail subband in all directions. uniform quantizing the approximation subband, non-uniformquantizing the detail subband, and combine the extracted multi-resolution co-occurrence matrix texture characteristics with statistical characteristics, the calculation of similarity between two images using the weight coefficient of evaluation formula.Finally this method was applied to standard Brodatz image library, all 112 of texture image for image retrieval.The experiment showed that this method can achieve better retrieval rate compared with the traditional wavelet.