A novel algorithm based on region of interest was presented for texture image retrieval. The proposed method was based on a combination of Curvelet multiresolution image decomposition and colour correlation histogram. First, Curvelet transform was applied for texture image multi-scale decomposition. Then, the region of interest was extracted based on the distribution of the energy of different scales, and a quantization step was used before computing autocorrelograms of the region of interest Curvelet coefficients. Finally, index vectors were constructed for similarity comparison. Experiments based on Brodatz benchmark database show that the algorithm is more effective compared to other retrieval methods based on Curvelet in texture image retrieval.

Key words：