Comparison of Texture Feature Extraction Method for COVID-19 Detection With Deep Learning

Abstract

This paper describes research on texture feature extraction for COVID-19 detection. Fractal Dimension Texture Analysis (FDTA) and Gray Level Co-occurrence Matrix (GLCM) were used for feature extraction. A dense neural network is used for classification. Three classes were used for classification to classify Normal, COVID-19, and Other pneumonia. The data entered in the texture feature extraction is a chest x-ray (CXR) image that is grey scaled and resized into 400x400 pixels. Performance analysis of the model uses a confusion matrix. The best performance feature extraction method for detecting COVID-19 is FDTA, with an accuracy testing of 62.5%.