Abstarct: Automatic segmentation of blood vessels from retinal image can help medical specialists for early diagnosis of many eye diseases. In this thesis we aim to propose an efficient algorithm for segmentation of blood vessel in colored retinal fundus image. Our algorithm contains three main stages: preprocessing, feature extraction and vessel construction. Preprocessing is the first stage which is divided into three steps itself. The Gamma correction, the wavelet transform and the modification of wavelet coefficients are these three steps. The second stage is extracting the centerlines using topographical features. In order to extract centerlines we use one of the twelve topographic labels namely ridge labels. It is found that ridge label properly matches the central points of the vessel. The extracted central points must be connected to produce the string of center lines. Further extension and growth of extracted points retrieves weak vessels from the image. For this purpose a directional filter bank is used to find the best direction along which central points can growth. Last stage is reconstruction of the vessels which gives a good approximation for width of vessel as each point. In many medical diagnosis the vessel widths is very important. The centerlines are used as seed points in a region growing procedure to reconstruct vessels. For this purpose the Top-Hat morphological operator is used to create four images by using circular structuring elements at different scales to prominent vessels with different width. These four images are compared against two predefined thresholds. The region growing approach with restriction to four binary images is used to reconstruct the vessel. The result of experiments reveals that the proposed algorithm works properly for both normal and abnormal retinal images.