K-core graph-based retinal vascular registration
Observing the differences of the blood capillary at different times is an effective method for contributing to the early diagnosis and treatment by differentiating the images. Therefore, distinguishing the differences of two retinal vascular images by registration should be regarded as the important precondition. We propose the method for solving the problem better than before. First a graph model of the vascular network is generated from the fundus image or other modality. Second, the graph is decomposed into a k-core representation which should transform a dense graph into a sparse version. Finally, the key nodes which kept high k-core value are remained for images registration by means of Iterative Closest Point algorithm. We gain the result that the average misalignment of k-core algorithm registration is infinitely close to zero, and the k-core algorithm is better than the SIFT algorithm clearly.