Abstract
Glaucoma is a condition of the eye in which optic nerve fibres are damaged that leads to loss of vision and may even lead to blindness if not diagnosed on time. Many manual scanning methods are available but they are time consuming expensive and require experts in these fields to use them. A number of automated glaucoma diagnosis methods using fundus images are also available. This paper compares the feature selection using Student’s t-test and Principal Component Analysis (PCA). It also proposes a modified decision making approach using bit-wise OR operation at the output of the classifier for different colour components. The R, G, B and grey scale values are extracted from the input image. This is then subject to 2D EWT to form the sub-band images. Then correntropy is extracted from these decomposed components. Features are selected using two methods: Student’s t-test and PCA. Features are classified using Least Squares Support Vector Machine (LS-SVM) to classify between normal and glaucoma images. For performance analysis, the impact of testing and training on the detection and performance analysis of the classifier was done. It is found that the accuracy of detection using Student’s t-test to the blue component is 78.8% and combined response of red, blue, green and grey component gives 97% whereas the accuracy of detection using PCA to the red component is 87.3% and the combined response of red, blue, green and grey component is 99%. With better detection accuracy, less processing time and computational complexity PCA was found to be better of the two methods.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have