A hybrid GLCM-CNN model for cataract detection in fundus image

Zulkernain, Syakirah Hanim (2023) A hybrid GLCM-CNN model for cataract detection in fundus image. Project Report. Melaka, Malaysia, Universiti Teknikal Malaysia Melaka. (Submitted)

	Text (24 Pages) A hybrid GLCM-CNN model for cataract detection in fundus image.pdf - Submitted Version Download (272kB)
	Text (Full Text) A hybrid GLCM-CNN model for cataract detection in fundus image.pdf - Submitted Version Restricted to Registered users only Download (5MB)

Abstract

Detection of cataracts based on fundus images is hard to perform due to errors in the image, such as uneven illumination, and some of the fundus images have low quality. The CNN single model can perform the detection of cataracts, but the misclassification is hard to reduce. This project proposes to use GLCM and CNN as feature extractors and DNN as a classifier to increase the accuracy of cataract detection while at the same time reducing the number of misclassifications of cataract and normal classes. The GLCM and CNN feature extraction are combined and become input for DNN for classification in performing cataract detection. The proposed model was built using PyCharm, and the results were visualised using classification reports, graphs, and a confusion matrix. The proposed model is compared with another two models, which are CNN and GLCM-k-NN. The performance results showed that GLCM-CNN can achieve high accuracy, which is 93%, but in terms of precision, sensitivity, and F1 score, they need to be enhanced for future works. The number of misclassifications still couldn’t be reduced as minimally as possible. The suggestion was made in this project for future work.

Actions (login required)

View Item