The vascular network of human eye is having variable size of blood vessels, contrast variations and outlier presence in pathological cases. For the better management of vascular diseases, automatic detection of vessel structures is required which is of great importance for its early diagnosis and treatment. A new model for blood vessel segmentation from fundus images based on stratified image matting is proposed. Generally, the image matting models requires an input trimap. The trimap generation is a time consuming task, also the accuracy of segmentation results depends upon the quality of input trimap. Therefore, a new method for generating a high quality trimap in less time, based on the application of kirsch template is incorporated in this image matting model, for accurate matte estimation. First, the good quality trimap is generated automatically by applying the Kirsch template which includes three parts: Blood Vessel Extraction Using Kirsch's Template, Co-fusion and Fuzzy C-Means Clustering. Then, for better segmentation performance, vessel pixels from unknown regions are extracted using the multi-level image matting model.

Keywords : Global Contrast Normalization (GCN), Conditional Random Field(CRF), Support Vector Machine, Patch Alignment Manifold Matting(PAMM), K-Nearest Neighbor, Fuzzy Multi-Criteria Evaluation (FMC).