An Efficient Local Binary Pattern Texture Descriptor Method for Quick Detection of COVID-19 Using a Deep Learning Algorithm

Abstract

This study has proposed an efficient local binary pattern (LBP) texture descriptor method for testing the patients with novel severe acute respiratory coronavirus syndrome (SARS-nCoV) and detecting COVID-19 using a deep learning algorithm from X-ray chest images. The proposed network-based architecture pretrained on the ImageNet is trained to collect X-ray images from the openly accessible databases and physicians. It is essential to choose an appropriate feature extraction technique for retrieving prominent, discriminating, and crucial information from the biomedical images for better efficacy. In this paper, after preprocessing and segmentation on X-ray chest images, a suitable feature descriptor (LBP) has been considered with the utmost care from the survey. LBP is known for its simplicity, versatility, distinct characteristics, and low computation complexity, leading to better performance. The proposed thesis works as network-based pretrained CNN architectures feature-based X-ray chest images. The experiment involved four-class classification models, such as Vgg16, ResNet50V2, Xception, InceptionResNetV2, which gave the best promising result. These images have been trained using the proposed LBP model and have achieved 91.7%, 93.85%, 97.6%, 97.7% for Vgg16, ResNet50V2, Xception, InceptionResNetV2, respectively, over 50 epochs and batch size 10.