An Investigation into Automated Retinal Image Segmentation

The thesis aims to investigate the performance and potential of wavelets in the context of feature detection in retinal imaging, and to use those results to develop a robust algorithm that can reliably detect the blood vessels within retinal photographs. Detection of blood vessels in retinal images is an important step in the predictions and diagnosis of cardio-vascular diseases, such as hypertension and diabetes, that are known to affect the appearance of the blood vessels in the retina. We analysed current image processing techniques for segmenting blood vessels from other features in retinal photographs. We considered an alternative approach that uses a model of the cross-sectional profile of the blood vessels, along with a series of noise filters, in an attempt to mimic how the human visual system would extract the vasculature from other features in retinal photographs. The results of the experiments demonstrate that the technique is able to detect blood vessels that are otherwise missed by conventional techniques, while rejecting fine details that are not blood vessels. Further work is needed to improve the results by refining the models and modifying their sensitivity to the borders of faint features in the image.