Novel angular binary pattern (NABP) and kernel based convolutional neural networks classifier for cataract detection

Cataract is considered as one of the foremost causes of blindness, especially among older people. In India, by the age 80, nearly half of older population either have cataract or they performed surgery for it. To avoid worse effects in eyesight like complete blindness or blurred vision, it is essential that cataract cases are detected in the initial stages for effective treatment. For detecting eye cataracts, the machines utilizing are exhibiting portability issues. Hence, this study is using digital image processing algorithms for the detection and classification of cataract on eye images along with its severity. Initially, the features such as color, shape and texture are extracted separately. Significantly, Novel Angular Binary Pattern- NABP is proposed for the texture feature extraction. The classification of images are performed in this study using the proposed Kernel Based Convolutional Neural Network after the feature extraction process. For all three feature types, the results are obtained separately. Performance of the proposed system is comparatively analyzed in terms of Accuracy, Sensitivity, Specificity, Precision, Recall and F – measure. In addition, comparative analysis is undertaken with respect to texture, colour and shape features. Classification results of the proposed system for varying epochs are also analyzed. Thus, all the analytical results confirmed the outstanding performance of the proposed system than conventional systems for cataract detection with 97.3% accuracy.