Classification of Cornel Arcus using Texture Features with Bayesian Regulation Back Propagation

Abstract

The corneal arcus (CA) is an eye problem frequently faced by some group of people. The CA signs indicate the presence of abnormal lipid in blood and can cause  several problems such as  blood pressure, diabetes, and hyperlipidemia. This paper presents a comparison of classification of the abnormal eye using a neural network. In order to extract the image features,  the gray level co-occurrence matrix (GLCM)was used. This matrix measures the texture of the image, where the statistical calculation can be used to present the image features. The Bayesian Regulation (BR) algorithm has been proposed, in which this classifier classifies the obtained results better than previous works by other researchers. In this experiment, two classes data-set of the eye image, normal and abnormal images CA are used. The results from this BR classifier demonstrate a sensitivity of 96.1 % and a specificity of 98.6 %. The overall accuracy of this proposed system is 97.6 %. Although this classifier does not obtain 100 % accuracy, however its result is  proven to be able to classify the CA images successfully.