Region-based Partial-Duplicate Image Retrieval

The spring up of large numbers of partial-duplicate images on the internet brings a new challenge to the image retrieval systems. Rather than taking the image as a whole, researchers bundle the local visual words by MSER detector into groups and add simple relative ordering geometric constraint to the bundles. Experiments show that bundled features become much more discriminative than single feature. However, in order to achieve more discriminative power, the bundle should be carefully extracted with elaborate contour, which can separate different objects and exclude irrelative visual words. In this paper, we propose a novel partial-duplicate image retrieval scheme based on regions extract by saliency analysis. In addition, to increase the discriminative power and robustness for coping with various transformations such as flip and rotation, the visual words detected in the region are further restrained by an affine invariant constraint. The affine invariant constraint employs the area ratio invariance property of affine transformation to build the affine invariant matrix for bundled visual words. Experimental results on the internet partial-duplicate image database verify the effectiveness of our approach.