A novel local descriptor based on image patch gray-value coding

Local descriptors computed for key-points or interest regions are successfully applied in computer vision area. Among various descriptors, the SIFT based descriptors have been demonstrated to perform best. However, their computational cost is very expensive. In the paper, we propose a fast local descriptor, which is composed of the coding computed from the image patch centered at each key-point. The new descriptor is easily constructed and has fewer dimensions, which extremely speeds up the representing process and also benefits the later matching process. The experimental results show that the representation of our descriptor is about eighteen times faster and the matching is about 1.5 times faster than the standard SIFT descriptor, while the performance decreases only slightly.