A SIFT descriptor with local kernel color histograms

SIFT (Scale Invariant Feature Transform) has proved to be the most robust local invariant feature descriptor in object recognition and matching. Being designed mainly for the gray images, SIFT shows its vulnerability when deal with color images. To overcome this problem and increase the descriptor´s distinctiveness, we introduce a new descriptor, a combination of the SIFT approach and the improved local kernel color histograms, which shows a better performance than the original SIFT through experiments. Moreover, the experiments results show that the radio of correct matches increase and the mismatch radio remain constant simultaneously.