Adaptive registration algorithm of color images based on SURF

Abstract

SURF (Speeded-Up Robust Features) overcomes the expensive computation of SIFT (Scale Invariant Feature Transform), but both of them are designed mainly for gray-scale images. To overcome this drawback, color information could be added in the registration. This paper presents a novel adaptive registration algorithm of color images based on the SURF descriptor. Firstly, key points and descriptors are calculated using SURF. Secondly, color information is superimposed upon the key points’ descriptor to construct a modified descriptor. Finally, a two-way matching method based on adaptive Euclidean nearest/second nearest discrimination law is designed for image matching. Extensive experimental evaluations show that the algorithm not only inherits the superior performance of the SURF algorithm, but also increases adaptive matching capability for color images.