Abstract: Feature extraction is one of the key steps in content based image retrieval,but one feature based approach expresses only partial attributes of an image, which unilaterally describes the image content and is short of enough resolving power. This approach may not achieve ideal result in the case of images varying greatly. An approach was proposed based on image color and texture features, wherein color moments is the color feature and the variances of the multi scale high frequency sub-bands of wavclet domain. In experiment section, the presented method was compared to the traditional histogram and color moments methods. Combining the two features for image retrieval, we analysed the weight distribution between the two features by using ANMRR evaluation method which MPEG-7 recommends. The experiment shows that the approach has good retrieval performance.

Key words: Content based image retrieval,Wavelet transform,Multi-scale analysis,Color moments