Robust Wide Baseline Point Matching Based on Scale Invariant Feature Descriptor

Abstract

In order to obtain a large number of correct matches with high accuracy, this article proposes a robust wide baseline point matching method, which is based on Scott's proximity matrix and uses the scale invariant feature transform (SIFT). First, the distance between SIFT features is included in the equations of the proximity matrix to measure the similarity between two feature points; then the normalized cross correlation (NCC) used in Scott's method, which has been modified with adaptive scale and orientation, is used to put more weight on the correct pairs in the matrix since the SIFT feature is only invariant to linear changes of light. The proposed method removes all the proximity information about the distance between feature points' locations in the Scott's method, which causes mismatch in wide baseline matching. Experimental results show that the proposed algorithm is invariant to changes of scale, rotation, light, and thereby provides a new effective way for wide baseline matching.