3D Polynomial Interpolation Based Local Binary Descriptor

Many efforts are devoted to develop binary descriptors due to their low complexity, and flexibility in case of embedded systems. Almost all works on binary descriptor conception didn’t exploit all information of a given patch; they just involved pixels intensities into binary test process. This kind of solution lack efficiency on patch description. In this paper, we propose to design a new descriptor based on 3D polynomial interpolation by used pixels intensities. We must take into account geometric positions of pixels. We suggest to divide the patch into equal grid cells (sub patches). Each sub patch undergoes a dimension augmentation. It becomes a 3-dimensional vector by considering intensities values as the third dimension. Based on 3D polynomial interpolation, we approximate the point cloud by a surface. This step is followed by a binary tests between all coefficients of polynomials situated in neighborhoods. Our method shows a considerable discrimination in case of high similarity. The results of our approach are evaluated on a well-known benchmark dataset exhibit a considerable robustness and reliability in front of severe changes. A computation costing is reported in the end of results section.