TOCSAC: TOpology Constraint SAmple Consensus for fast and reliable feature correspondence

Abstract

This paper aims at outliers screening for the feature correspondence in image matching. A novel robust matching method, called topology constraint sample consensus (TOCSAC), is proposed to speed up the matching process while keeping the matching accuracy. The TOCSAC method comprises of two parts, the first of which is the constraint of points order, which should be invariant to scale, rotation and view point change. The second one is a constraint of affine invariant vector, which is also used to validate in similar and affine transforms. Comparing to the classical algorithms, such as RANSAC (random sample consensus) and PROSAC (progressive sample consensus), the proposed TOCSAC can significantly reduce time cost and improve the performance for wide base-line image correspondence. © 2009 Springer-Verlag.