High-performance recognition of corneas and crystalline lens from highly noisy images

Abstract

This paper proposes an approach to human-eye cornea and crystalline lens edges acceptance in digital images, when captured in environments that returns highly-noisy images, in which edge-detection algorithms do not perform very well. The proposed approach is supported by previously known properties of the cornea and the crystalline, beyond specific filters based on such properties. After being computationally implemented, these properties are imposed on the set of the acquired data, allowing conclusions about what data is correct, wrong and possibly wrong. While the stated properties will always produce true results, the filters can cause data losses, and may thus be implemented depending on the context and on the desired accuracy. By comparing this approach with the conventional one, which relies on suitable digital filters, there were observed large speed-ups, for relatively equal final detections.