COVID-19 is a new pulmonary disease that has been straining the global healthcare system because of its high occurrence. It has been found that early-stage COVID-19 can be diagnosed using chest X-ay (CXR) images. Till now, most of the research has concentrated solely on the application of deep learning algorithms, which are valuable but lack proper pre-processing of CXR images. In this context, the purpose of this work is to study the cumulative effects of enhancement approaches on the performance of deep learning models. Within this research, four distinct iterations of histogram equalization image enhancement techniques were utilized on the chest X-ray (CXR) images. These encompass Median-Mean based Sub-Image Histogram Equalization (MMSICHE), Exposure based Sub-Image Histogram Equalization (ESIHE), Dominant Orientation based Texture Histogram Equalization (DOTHE), and Edge-based Texture Histogram Equalization (ETHE). The improved images are subsequently input into two pre-trained neural networks from the Visual Geometry Group (VGG) family, namely the VGG-16 and VGG-19 models, for the purposes of categorizing the CXR images into three categories: COVID-19, normal, and pneumonia. Ultimately, it was observed that the VGG-16 model employing the ESIHE image enhancement technique yielded the highest accuracy, reaching 92.17%.
