Glaucoma Detection using HOG and Feed-forward Neural Network

Abstract

The second most common cause of blindness in the world is glaucoma. It is an eye condition that doesn't show any early symptoms until considerable vision loss has already taken place. To reduce the risk of vision loss, glaucoma diagnosis is therefore crucial. This study provides an assessment of detecting retinopathy, glaucoma, and healthy images using a feature extraction method called Histogram of Oriented Gradients (HOG), and feed-forward neural network used as a classifier to classify the refunds images. The Contrast Limited Adaptive Histogram Equalization (CLAHE) method, morphology-based homomorphic filter, wavelet-based homomorphic filter, andmulti-scale top-hat transformation can all be used to enhance retinal images and increase their quality and dynamic range. Retinal fundus images are enhanced to increase contrast and bring out the retinal vessels. From the HOG we extracted the different mean, variance, Skewness, kurtosis, entropy, and energyfeatures set. Further, the extracted features set are given to the feedforward neural networks to classify the images.