Abstract: Image retrieval based on the texture and shape in the DCT compressed domain is proposed by introducing the spatial information of DCT blocks and DCT coefficients into the feature extraction. First, the definition of complex is introduced. Then, according to the complex of the DCT blocks and distribution of DCT coefficients, a weighted complex histogram is presented to extract the texture information. At the same time, DCT blocks are classified into different edges using the DCT coefficients directly and the edge-spatial distribution feature is adopted to describe the image shape. The statistical distribution of DCT blocks and the spatial distribution of DCT coefficients are taken into account in this paper, thus avoiding the wrong retrieval and missing retrieval. Experimental results show that the new method has a good performance in both retrieval efficiency and effectiveness.

Key words: content-based image retrieval, discrete cosine transform(DCT), weighted complex historgram, edge-spatial distribution feature