X-Ray Image Classification Using Support Vector Machine Based on Histogram of Oriented Gradients and Local Binary Patterns Feature Extraction

Abstract

This study aims to compare the effectiveness of two feature extraction methods, namely Histogram of Oriented Gradients (HOG) and Local Binary Patterns (LBP), in classifying X-ray images using Support Vector Machine (SVM) as the classification algorithm. In this research, the X-ray images were first processed through a preprocessing stage to enhance image quality. The features of each image were then extracted using the HOG and LBP methods, which were subsequently used as input for the SVM model to classify the X-ray images. We used a 3500 dataset with a balanced distribution of 700 images for each of the five classes, which are Normal, Covid-19, Pneumonia-Bacterial, Pneumonia-Viral, and Tuberculosis. The data was divided using the K-Fold Cross Validation method with 5 folds, where in each iteration, one fold was used as test data 20% (700 images), and the remaining four folds 80% (2800 images) were used as training data. The experiment was done in three scenarios to compare the effectiveness of three features: HOG, LBP, and a combination of HOG and LBP. The first experiment using the HOG feature has results as follows: average of accuracy is 84%, precision 84%, recall 84% and F1 score 84%. By using LBP, the performance of classification decreased with an average accuracy of 68%, precision 68%, recall 68% and F1 score 68%. The combination of HOG and LBP has a good result with an average accuracy of is 84%, precision 84%, recall 84% and F1 score 84%.