Novel image retrieval method based on interest points

Abstract

Traditional interest points based image retrieval methods employed local features of interest points to describe image. The difference between local areas of interest points and the region of interest limited the retrieval accuracy. Considering the shape characteristic of interest points, this paper presented a new image retrieval method based on region of interest determined by interest points. Firstly, interest points were detected by tracking wavelet coefficients of different scales. Secondly, the convex hull of interest points was calculated to extract the region of interest in the image. Then color and shape features of the region were used to describe an image. Finally, the weighted feature distance was used to define the similarity between two images. With robustness to image rotation, translation and scale, the method makes the retrieval implementation at the object level and avoids the shortcoming of traditional methods. Lots of experiments based on an image database containing 1100 images show that the method improves the average retrieval precision over 11.1 percent, compared with other interest points based retrieval methods.