A novel image retrieval method based on block truncation coding(BTC) is proposed to solve the problems of gray level co-occurrence matrix(GLCM).Firstly,the image is partitioned into non-overlap blocks of certain size.BTC is generated for each block independently,which can efficiently describe the image texture information,spatial distribution and shape feature.On the basis of this,the texture primitive is defined.Then,an improved co-occurrence matrix based on the texture primitive is developed to extract the texture and shape features for the image retrieval.Experimental results have shown that the proposed method has sound and robust retrieval performance by integrating spatial and shape information into image texture descriptors.