| In this thesis, a novel Perceptual Organization (PO) based approach was developed for blood vessel extraction from retinal images. In the method, an image is first segmented in real-time into an edge trace map in which each trace is an ordered list of edge pixels. Each edge trace is divided into perceptual features of Generic Curve Segments (GCSs) at Curve Partitioning Points (CPPs) based on PO criteria. Accordingly, a vessel map is described by GCSs (vessel segment features) and CPPs (junction features). Each vessel segment consists of two associated parallel GCSs. Junctions, which can be defined using CPPs, are the joinings of some vessel segments and were detected by a matched filter method. A vessel map is extracted by integrating the identified vessel segments and junctions. The experiments were conducted in comparing with the latest benchmark system. The improvements include both the effectiveness and the efficiency of detecting vessels. The results demonstrate a great potential for PO based methods. (Abstract shortened by UMI.)... |
