Abstract: Aiming at the problem of many error matching points and slow matching speed of SIFT algorithm, this paper presents a SIFT pixel filter pretreatment dimensionality reduction bidirectional matching method. Before SIFT constructing DoG space, the pixel point needs to preprocess, in order to reduce the useless feature points producing. During feature descriptor generating, the dimensionality needs to be reduced, in order to reduce the amount of computation. Finally, using the constraint registration method, the SIFT bidirectional matching is realized. The experimental results show that the improved method can improve the accuracy and efficiency of matches.

Key words: SIFT feature matching, pixel filter, dimension reduction, bidirectional matching