Retinal vessel enhancement via sparse coding and dictionary learning

Abstract

This paper presents a new method for vessel enhancement in retinal fundus images. In the proposed method, two dictionaries are utilized to gain the vessel structures, including Representation Dictionary (RD) which is generated from the original vessel images and Enhancement Dictionary (ED) generated from the corresponding label images. Then sparse coding technology is utilized to represent the target vessel image. At last, the learned dictionaries is used in the enhancement process. However, the size of the dictionaries increases the computational burden on the sparse coding process, which implies sophisticated data management and memory access. Here we introduce a simplified formulation that reduces the size of the dictionaries. Besides visual validation, quantitative evaluations demonstrate that our strategy is able to yield results that are highly competitive with traditional approaches.