Automated analysis of retinal images

摘要

We describe a method for segmenting retinal images into irregularly shaped primary regions which are bounded by vessels and other structures. Boundaries are classified using grey level models. We define a process of merging primary regions into large patches using image properties such as texture and intensity, and semantic interpretations of boundaries and their measured properties. The degree to which each merged region is perfused by microcapillaries is assessed using a texture measurement. The method which makes extensive use of morphological processing depends on a limited number of parameters which have natural physical interpretations. We obtain good agreements with expert analyses of the images.