Multiscale Gabor Wavelet for Shoeprint Image Retrieval

Abstract

The problem that the shoeprints found at crime scenes are often poor quality and with lots of noise causes much trouble when extracting the basic shapes and later classifying the database of shoeprint images automatically. In this paper we present a shoeprint image retrieval method based on odd and even Gabor filter. The method combines odd and even Gabor filter to extract the texture and geometry features and suppress noise after wavelet package decomposition. The texture features saved in a tree are used to query and the geometry features are used to weight the similarity. Experimental results demonstrate that the proposed method is robust and effective to match the incomplete and noisy shoeprint images.