Retinal image registration based on bifurcation point and SURF

Abstract

Retinal image registration is the process of matching and superimposing two retinal images of the same patient. The traditional feature-based retinal image registration algorithm is computationally expensive during the matching process. This paper proposes a fast and efficient registration method based on the combination of bifurcation point and SURF algorithm. First, the eight-neighbor search algorithm is used to detect the bifurcation points of the reference image and the target image, and then the SURF feature is extracted in the rectangular template region centered on the bifurcation point. The Euclidean distance is used to perform rough matching on the extracted features, then RANSAC is used for fine matching, and finally the transformation model is estimated. Experiments show that this method can quickly and effectively achieve the registration of retinal images while reducing a large number of unnecessary searches and achieving a great registration result.