Segmentation of Blood Vessels in Retinal Fundus Images for Early Detection of Retinal Disorders: Issues and Challenges

Retinal disorders are progressive in nature and remain passive for years together without causing any visual indication of disorder even to the subject themselves. Hence automated and intelligent methods of analysis of retinal scanned images are quite necessary to improve accuracy and detection time to aid in early detection and consequent treatment. This paper presents the findings of a vast literature survey done with respect to automated detection techniques by analyzing their underlying principles and obtained performance results. The entire survey has been done based on two main evaluation metrics namely detection accuracy and time of convergence. This is based on the underlying principle that migration from manual and conventional methods to automated systems is to improve the accuracy by overcoming the errors incurred in manual detection methods and at the same time to reduce the painstakingly long time required in the manual method of observation and detection. This paper proposes a computation complexity reduction mechanism in dehazing by utilizing the convolution properties of deep belief neural networks to train the data sets in the least possible time with improved image quality.