A new SIFT keypoint descriptor for copy detection

摘要

In this paper, we propose a novel keypoint descriptor coined Spatial Coherent Feature (SCF) based on Scale Invariant Features Transform (SIFT) and the spatial coherent information. Like SIFT, our descriptors encode the salient aspects of image gradient in the keypoint's neighborhood. However, instead of only using 4*4 sample regions to computer orientation histogram, we get descriptor based on eight consecutive neighborhood regions, and each region has different size but approximate numbers of pixel. That is to say, the new descriptor is computed on spatial consecutive regions, and it includes neighborhood information around keypoint. The performance of SCF descriptor is tested for copy detection. Experimental results demonstrated that the novel keypoint descriptor can be robust for some kinds of attacks such as scale, rotation and reduce the error matching because of introducing spatial coherency, compared to SIFT descriptor.