Subregion based local descriptor

A novel distinctive descriptor named MSOGH is proposed, which is able to well represent the interest region and is robust to photometric transformations and geometric transformations. According to intensity order, subregions are firstly constructed. Then feature descriptor of the subregion is computed by point permutation of the sample points in each subregion. Finally, feature descriptor of the region is formed by concatenating all subregion feature descriptors. The discriminative power of the proposed descriptor is compared with 5 major existing region descriptors (MROGH, SIFT, GLOH, PCA-SIFT and spin images). Extensive experimental results show that the proposed descriptor achieves better performance than state-of-the-art descriptors.