An Automatic Approach for Retinal Vessel Segmentation by Multi-Scale Morphology and Seed Point Tracking

We proposed a complete algorithm for enhancing and segmenting retinal vessel using multi-scale morphology and seed point tracking approach. First, the high contrast blood vessel images at each scale are obtained by the comprehensive application of top-hat and bottom-hat transformation enhancement technology with line structuring elements, the bright and dark areas, whose diameters are greater than the scale of the structuring element, can be filtered in this stage. Second, the blood vessel image is segmented by multi-threshold based vessel tracking technology. In the tracking stage, the thresholds are adaptively obtained using the proportion of the blood vessel pixels, and the stop condition can be automatically calculated in this process. The performance of our proposed method is assessed on the publicly available DRIVE and STARE fundus image datasets. For database DRIVE, the proposed method has achieved accuracy, specificity and sensitivity of 0.9449, 0.9810 and 0.7236 respectively, and 0.9460, 0.9680 and 0.7486 for database STARE respectively. The segmentation result of our novel algorithm is better than the other unsupervised methods. Our new technique is robust to the pathological cases, it improves the segmentation accuracy and decreases the false segmentation near the large bright and dark areas, such as optic disc, hard exudate, fovea and hemorrhage. The proposed approach is an unsupervised method and does not demand training phase. Furthermore, the method can be implemented efficiently and can be stopped automatically, the user interaction or adjustment of parameters is not necessary.