Contribution to image segmentation

Abstract

Retinal imaging is a powerful tool for detecting and diagnosing various health conditions. Locating retinal vessels is important because it allows for the specification of different tissues of the vascular structure. Ophthalmologists use images with binary segmentation of retinal fundus to analyze and predict diseases such as hypertension and diabetes. However, blood vessel segmentation in retinal images can be challenging due to several factors such as low contrast, background illumination inhomogeneity, and noise. The main objective of this dissertation is to study and propose effective methods for automatic retinal vasculature segmentation. Our first contribution involved studying the performance of improved FCM algorithms, including FCM, EnFCM, SFCM, FGFCM, FRFCM, DSFCM_N, FCM_SICM, and SSFCA, to recommend the best ones for the segmentation of retinal blood vessels. We evaluated their performance based on three criteria: noise robustness, blood vessel segmentation performance, and execution time. In our second contribution, we proposed a new unsupervised method that ensures high-accuracy detection compared to previous studies. It depends on hybrid filtering and adaptive thresholding. We validated our proposed studies using two benchmark databases: STARE and DRIVE. This dissertation also includes contributions related to the segmentation of MR brain images to identify tumors and different tissues. These contributions involve the development of new methods that have been evaluated using various databases and have shown promising results. These contributions are included in the appendices of the dissertation. Overall, this dissertation aims to contribute to the field of medical image segmentation by proposing effective new methods, which can help in disease detection and monitoring.