| Summary: Retinal blood vessel segmentation is important for diagnosis and treatment of various pathologies such as hypertension, diabetes and cardiovascular diseases. In this thesis, blood vessels in color retinal fundus images are segmented automatically and then a characteristic feature matrix is extracted in order to identify segmented binary blood vessel network. Blood vessel segmentation consists of four stages: 1) Preprocessing, 2) Blood vessel enhancement, 3) Pixel based classification and 4) Post-processing. Firstly, the retinal region is selected before retinal region expansion, gray level transform and vessel light reflex removal processes are implemented in the preprocessing stage. Two-dimensional Gauss and Gabor filter and Frangi filter are applied separately before morphological top-hat transform which extracts details from an image as blood vessel enhancement methods. In the pixel-based classification stage, rule-based methods using thresholding, unsupervised methods using clustering and supervised classification approaches using machine learning methods are implemented. Some statistical and color-based features are also used with the supervised classification methods. Afterwards, post-processing methods are applied to binary vessels and performance evaluation is performed. In the last stage of the thesis, skeleton of the binary vessels are obtained before feature points extraction and a characteristic feature matrix is structured in order to use for image registration as a future process.
Keywords: Retinal images, Blood vessel segmentation, Vessel enhancement, Thresholding, Clustering, Supervised classification, Feature extraction, Matched filters |
