Color-SURF: A surf descriptor with local kernel color histograms

Abstract

SIFT (Scale Invariant Feature Transform) is an important local invariant feature descriptor. Since its expensive computation, SURF (Speeded-Up Robust Features) is proposed. Both of them are designed mainly for gray images. However, color provides valuable information in object description and matching tasks. To overcome the drawback and to increase the descriptor's distinctiveness, this paper presents a novel feature descriptor which combines local kernel color histograms and Haar wavelet responses to construct the feature vector. So the descriptor is a two elements vector. In image matching process, SURF descriptor is first compared, then the unmatched points are computed by Bhattacharyya distance between their local kernel color histograms. Extensive experimental evaluations show that the method has better robustness than the original SURF. The ratio of correct matches is increased by about 8.9% in the given dataset.