To Örebro University

Bifurcation points are typically considered as landmark points for retinal image registration. Robust detection, description and accurate matching of landmark points between images are crucial for successful registration of image pairs. This paper introduces a novel descriptor named Binary Descriptor for Retinal Bifurcation Point (BDRBP), so that bifurcation point can be described and matched more accurately. BDRBP uses four patterns that are reminiscent of Haar basis function. It relies on pixel intensity difference among groups of pixels within a patch centering on the bifurcation point to form a binary string. This binary string is the descriptor. Experiments are conducted on publicly available retinal image registration dataset named FIRE. The proposed descriptor has been compared with the state-of-the art Li Chen et al.’s method for bifurcation point description. Experiments show that bifurcation points can be described and matched with an accuracy of 86–90% with BDRBP, whereas, for Li Chen et al.’s method the accuracy is 43–78%.