Segmentation of vessels in retinal images by shortest path histogramming

The analysis of ocular fundus images is important for the detection of disease, tracking changes in the retina over time, and 3D retinal reconstruction. Blood vessel analysis plays a major role in such processing. We propose to use a shortest path analysis to identify blood vessels in retinal images. Ideally, the shortest path between every pair of pixels (where speed through a pixel is proportional to the gray level) would be found, and then a counter incremented for every pixel on the path. Once all pixel paths through a pixel have been counted, the highest counts should correspond to vessel pixels. However, this has exponential complexity in the number of pixels. Thus, not all pixel pairs can be analyzed. We investigate here the selection of a small subset of pixels to seed the process, and propose two methods: (1) the curl method, and (2) the ID degree 2 polynomial method.