Context-Dependent Segmentation of Retinal Blood Vessels Using Hidden Markov Models

Hidden Markov Models (HMMs) have proven valuable in segmentation of brain MR images. Here, a combination of HMMs-based segmentation and morphological and spatial image processing techniques is proposed for the segmentation of retinal blood vessels in optic fundus images. First the image is smoothed and the result is subtracted from the green channel image to reduce the background variations. After a simple gray-level stretching, aimed to enhance the contrast of the image, the feature vectors are extracted. The feature vector of a pixel is formed from the gray-level intensity of that pixel and those of its neighbors in a predefined neighborhood. The ability of the HHMs to build knowledge about the transitions of the elements of the feature vectors is exploited here for the classification of the vectors. The performance of the algorithm is tested on the DRIVE database and is comparable with those of the previous works.