Performance Evaluation of LBP, Mutual Information and CNN in Digital Diseases Image Detection

Digital image processing is a field that employed computer algorithms to analyze digital images. Several images have been subjected to digital classifications like trauma-related, hypertensive, cancer, plant leaf diseases, breast cancer, etc. Feature extraction/selection is a pre-processing technique that removes redundant features from images. Several feature extraction/selection techniques, especially statistical and deep learning methods, have been employed by researchers but their performances have not dealt with adequately. This work focused on the performance comparison of three selected feature extraction and selection techniques viz; Local Binary Pattern, Mutual Information and Convolutional neural network in digital image processing. The datasets of MRI brain tumour images from the Kaggle website of 394 were pre-processed and also subjected to feature extraction and selection using the selected techniques. The extracted features were classified by Support Vector Machine and the outcome were evaluated by confusion matrix parameters.. The results showed the CNN-SVM based Image detection system at threshold of 0.85 produced Recall 84.8%, Specificity 95.9%, False Positive Rate 4.2%, Accuracy 92.9% and Precision 88.1%; the MI-SVM system produced Recall 70.5%, Specificity 92.0%, False Positive Rate 8.0%, Accuracy 86.3% and Precision 76.3%; while LBP- SVM system produced Recall 64.8%, Specificity 88.9%, False Positive Rate 11.1%, Accuracy 82.5% and Precision 68.0%.