RSD-HoG: A new image descriptor

Abstract

In this paper we propose a novel local image descriptor called RSD-HoG. For each pixel in a given support region around a key-point, we extract the rotation signal descriptor(RSD) by spinning a filter made of oriented anisotropic half-gaussian derivative convolution kernel. The obtained signal has extremums at different orientations of the filter. These characteristics are combined with a HoG technique, to obtain a novel descriptor RSD-HoG. The obtained descriptor has rich, discriminative set of local information related to the curvature of the image surface. With these rich set of features, our descriptor finds applications in various branches of computer vision. For evaluation, we have used the standard Oxford data set which has rotation, brightness, illumination, compression and viewpoint changes. Extensive experiments on these images demonstrates that our approach performs better than many state of the art descriptors such as SIFT, GLOH, DAISY and PCA-SIFT.