Quadratic filter for the enhancement of edges in retinal images for the efficient detection and localization of diabetic retinopathy

This paper describes the design and implementation of efficient edge detection quadratic filters for better localization of microaneurysms, caused by diabetic retinopathy, in fundus retinal images. The method is based on Volterra filter that accounts for majority of polynomial nonlinearities in images. Teager filters are designed and implemented for detecting edges in retinal images generated by a fundus camera. Better localization of microaneurysms is achieved with an isotropic quadratic filter whose kernel is designed based on optimization. The noise performance of the edge detectors is tested with Gaussian and impulsive noise.