My thesis consist of three aspects: 1. In order to extract the low contrast vessels in retinal images, we propose a novel enhancement algorithm based on directional field and develop a robust retinal vessel extraction framework. The proposed enhancement algorithm can significantly improve the contrast of vasculature especially the slim vessels and low contrast vessels. Then a hybrid extraction algorithm is applied to extract the vasculature. 2. Quantification of retinal vessel change is difficult and complicated because of the width variation and the local contrast unsteadiness and the partial volume effect. We propose a new method for re-segmentation of retinal vessels in fundus photographs that uses Neighbor Function to reclassify the vessel edges. 3. Detection of vascular bifurcations is a challenging problem in multimodal retinal image analysis. Existing algorithms using bifurcations as cues usually fail in correctly aligning poor quality retinal image pairs. To solve this problem, we propose a novel highly distinctive local feature descriptor named Partial Intensity Invariant Feature Descriptor (PIIFD), and describe a robust and automatic retinal image registration framework named Harris-PIIFD. In our framework, we first used corner points instead of bifurcations as control point candidates because corner points are sufficient and uniformly distributed across the image domain. Secondly, PIIFDs were extracted for all corner points following a bilateral matching technique to identify corresponding PIIFDs matches between image pairs. Thirdly, any incorrect matches were removed and any inaccurate matches refined. Finally, an adaptive transformation was used to register the image pairs.
修改评论