Latticed Corner Detector Driven by Best-Fit Differentials in Arbitrary Rational Directions

In this paper, we suggest a new corner detector based on latticed differentials in arbitrary rational directions and propose a response-suppression in the neighborhood of an extreme. The latticed differentials are defined by a Direction let-like decomposition (DLD, we also explain it as differential on latticed decomposition) to exclude the pseudo corners due to unsmooth discrete sampling along image edges and in homogeneous regions. Besides, a response-suppression procedure is proposed to reduce the repetitive detection of corners in a small neighborhood, which is resulted by the higher differential values around a maximum or a minimum point. Compared with GLCP and FAST methods, combined DLD method has 5% increase of the average accuracy values (ACU) and prevents false detected corners effectively.