B-SIFT: A highly efficient binary SIFT descriptor for invariant feature correspondence

Abstract

This paper presents a novel binary descriptor named as B-SIFT(Binarized Scale Invariant Feature Transform) for efficient invariant feature correspondence.Through analyzing the local distinctive gradient pattern, we convert the standard SIFT descriptor to a binary representation which can be computed extremely fast with bitwise operation. Extensive correspondence trials based on a benchmark Oxford image data set with viewpoint, scale, image blur, JPEG compression and illumination changes demonstrate that in general, the proposed B-SIFT method significantly outperforms the standard SIFT with over 400 times faster in matching time and 32 times less in memory resources, while achieves the same matching score as SIFT. © 2012 Springer-Verlag.