Image Retrieval Using Visually Significant Points

To solve the question of the traditional points in image retrieval, a robust and self-adaptive point extraction algorithm is proposed in this paper. The block difference of inverse probabilities (BDIP) image is firstly obtained from the original image based on the BDIP model. According to the distribution feature of the BDIP image, the local points are extracted. Then, the color-spatial feature surrounding the local points is extracted as an index for image retrieval. Comparisons are conducted between this algorithm and other descriptors. Experimental results show that our method provides more significantly retrieval results than the traditional ones.