A FAST INTEREST POINTS MATCHING METHOD BASED ON RETINEX MODEL

摘要

The classical algorithm of SIFT (Scale Invariant Feature Transform) has poor real-time in the process of interest points matching. A fast interest points matching method based on biologically inspired retinex model is proposed. Firstly, the image details enhancement and color constancy are implemented with the Retinex model. Then the algorithm of CCSIFT (Color Constancy Scale Invariant Feature Transform) consist of SIFT and gray value is extracted from processed integral image, and it has a good invariance in image scaling, rotation, translation and robust across a substantial range of affine distortion. Interest points are matched according to gray-scale-first search strategy which can reduce the computational cost largely. The results of experiments with university of Southern California scene database show that the recognition time of the proposed fast method is reduced two orders of magnitude compared with the classical SIFT algorithm, while guaranteeing the recognition rate, so scene classification of such near real-time application using it become possible.