Abstract:
Ophthalmologists rely on retinal vessel segmentation. Low contrast, variable vessel size and thickness, and micro-aneurysms and hemorrhages complicate this issue. Early methods used hand-crafted filters and morphological post-processing to capture vessel structures.
Deep learning has improved segmentation accuracy recently. We propose a domain-enriched deep network with two components: 1) a representation network that learns geometric features specific to retinal images, and 2) a computationally efficient residual task network that uses these features to segment pixels.
Any training set learns representation and task networks together. We propose two new constraints inspired by expected prior structure on representation filters to obtain physically meaningful and practically effective filters: 2) a data adaptive noise regularizer that penalizes false positives.
Thin vessel detection is improved with multi-scale extensions. The proposed prior guided deep network outperforms state-of-the-art alternatives in common evaluation metrics while being smaller and faster.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
