Retinal Blood Vessel Segmentation Using Phase Congruency Model

Retinal blood vessels are the only part of blood circulation that can be observed directly. Doctors manually segment retinal images to evaluate the progression of diseases like diabetes and hypertension. The manual tracking is a tedious process. The retinal images have low and uneven contrast with respect to the background. Thus, a robust model is needed for describing the intensity variations for segmenting the vessels. Phase congruency based edge detector using the principal moments is very well adapted to this description. The paper purposes a simple contrast invariant retinal image segmentation using phase congruency model with a good edge localization. Initial results using STARE database (average sensitivity of 0.94, specificity of 0.66 and average accuracy of 0.91) are promising and comparable with other techniques in literature.