A biometric identification algorithm based on retinal blood vessels segmentation using watershed transformation

A biometric system establishes an authenticity of a specific physiological or behavioral trait of an individual. In this paper, we evaluate a retinal identification algorithm based on fundus images mainly on retinal vascular network that is a characteristic of the most reliable biometric identification. In order to extract the features, a segmentation of the vascular network is performed using a powerful morphological technique called watershed. This technique allows extracting, faithfully, the vascular skeleton that will eventually be used for detecting biometric attributes of the network such as bifurcation points and crossing branches. Finally, our algorithm performs model comparison made based on these characteristics. We test our algorithm on a retinal images database ARIA. The experimental results are interpreted and a decision threshold of the correspondence between the images is determined. For evaluating, the resulting classification system is tested on the images; its sensitivity and specificity are then estimated.