Automatic retinal image registration based on blood vessels feature point

Abstract

Retinal image registration is commonly required in order to combine the complementary information in different retinal modalities images, which is very useful for clinic diagnoses and laser treatment. In the paper, an automatic scheme to register retinal images based on blood vessel feature point extraction is presented. Firstly, the crossover and branching points of the RF (red-free) retinal image and the FA (fluoroscein angiography) retinal image are extracted using a multi-direction exploratory method after the blood vessel visibility is enhanced. Considering the number of feature points extracted on different images are not equal and their location does not correspond between the retinal images, the scheme described in the paper combined the closest interval criterion between the corresponding feature points with a SVD algorithm to accurately register retinal images. Our method compared with previously known ways is an efficient and automatic scheme of retinal image registration. The experimental results indicate the registration of the fluoroscein angiography image with the corresponding red-free retinal image and a series of the fluoroscein angiography images can be also successful using our method.