Retina is one of the key features of the human eye. Analysis of retinal images is essential for early detection and diagnosis of pathologies. This thesis deals with retinal image processing and analysis of colour fundus images. The main contributions of this thesis are: • A geometrical and morphological analysis of the retinal vasculature. Starting from a binary image of the vascular network, the analogy between the vasculature and the structure of a graph was explored, yielding a matrix representation of blood vessels. Properties of each graph edge (corresponding to vessels) were computed, such as length, width, area, curvature, branching angles. This analysis resulted in developing a novel tortuosity measure taking width information into account, as well as a novel approach to the detection of the retinal main vein. • Context-based detection of a reference frame of the retina, by detecting the mai'n retinal landmarks: the optic disk, the macula, an approximation of the main blood vessel. Our detection algorithms introduce a technique based on plausible detection and constraint satisfaction, which provides several optic disk candidates among the brightest image regions, and several macula candidates among the darkest image regions, and fits two half parabolas centered at the optic disk candidate by minimizing a cost function. The actual landmarks are chosen among the candidate landmarks that satisfy several constraints imposed by the anatomy of the retina. • A preliminary architecture for a fuzzy diagnosis system for retinal pathologies. The system inputs are several symptoms related to a certain disease, and the output is a grading of the disease.An example of a fuzzy diagnosis system was presented for diabetic retinopathy. Keywords: image processing, retina, optic disk, macula, main vessel, tortuosity, diagnosis, fuzzy, main vein, vascular morphology.