ABSTRACT
CT image diagnosis of COVID-19, an infectious disease that causes respiratory problems, proved efficient with CNN-based methods. The accuracy of these machine learning methods relies on the quality and dispersion of the training set, which has often been ensured by utilizing the preprocessing strategies. However, few studies investigated the impact of different preprocessing methods on accuracy rates in diagnosing COVID-19. As a result, a comparative study on different image preprocessing methods was done in this work. Two popular preprocessing methods contrast limited adaptive histogram equalization (CLAHE) and Discrete Cosine Transform (DCT), which were processed and compared in a CNN-based diagnosis framework. With a mixed and open-source dataset, the experimental results showed that DCT based preprocessing method had a higher accuracy on the test set, which was 92.71%. © 2021 IEEE.