A voting procedure supported by a neural validity classifier for optic disk detection

Abstract

In this work a Voting Procedure supported by a Neural Validity Classifier for assuring a correct localization of the reference point of optic disk in retinal imaging is proposed. A multiple procedure with multiple resulting points is briefly described. A Neural Network behaving as a Validity Classifier of regular/abnormal solutions is then synthesized to validate the adequacy of the resulting midpoints as candidate reference points. A Voting Procedure, supported by the synthesized Neural Validity Classifier, is successively performed, by comparing only candidate pixels classified as valid ones. In this way, the most suitable and reliable candidate can be voted to be adopted as the reference point of OD in successive retinal analyses. © 2012 Springer-Verlag.