Design And Implementation Of Contourlet Texture Image Retrieval System

The development of multimedia technology gradually makes the image database resources become more and more abundant. How to quickly and accurately retrieve the relevant images has become an urgent problem in the image database. In order to solve this problem, texture image retrieval system based on content has been proposed and widely used. Texture image retrieval system has two main components of which are feature extraction and similarity measure. Therefore, in order to improve the performance of retrieval systems, we need do the appropriate research in the feature extraction and similarity measure aspects.Compared with wavelet, Contourlet transform is a distinct representation in the image multistate geometric analysis. Subband coefficients statistical parameters were researched on the texture image retrieval system in the Contourlet transform, which used to improve texture retrieval performance. This paper compares the mean, variance, the absolute value of the mean, mean energy, skewness and kurtosis. Canberra distance is the system similarity function. The experiment proved that absolute value of the mean could improve the retrieval rate in the single feature under the same length of feature vectors, retrieval time and storage space; Through mathematical theory and experiments showed that image texture feature vector were extracted with cascaded energy and peak has a higher retrieval rate than the same structure Contourlet retrieval method. The greatest influence features R3 in the single feature of texture> the greatest influence features R36 in the double feature of texture and the greatest influence features R362, R3641, R634512 in the more features of texture were analyzed, no matter how the decomposition parameters and the use of the filters changes, other statistical parameters were approximately the same in the Contourlet texture retrieval algorithm in addition to single feature had quite an effect on image texture system. Thinking of time and space resources, double feature was generally picked to use in image retrieval system.Real Contourlet Transform is the lack of phase information with a lower level of shift invariant. Thus Texture features are inaccuracies and comprehensive described. For this reason based on the complex Contourlet transform texture image retrieval system-related is proposed, but the kind of texture retrieval system retrieval time is too long. How to reduce texture image retrieval time? Which similarity measurement method is suitable to dual tree complex Contourlet domain. First, cascading subband amplitude and relative phase to structure vectors in dual tree complex Contourlet domain. And then the 109 texture image of the Brodatz databases were retrieved based on the Canberra distance, accordingly the image subsets similar to the query images was selected. The final retrieval rates of the whole system and retrieval time were calculated. Experimental results show that it is not only reduces the search time more than six times, but also increases the average rate of the texture image retrieval, and easy to implement, which improve the performance of texture image retrieval system.