COVID-19 digital x-rays forgery classification model using deep learning

Abstract

Nowadays, the internet has become a typical medium for sharing digital
images through web applications or social media and there was a rise in
concerns about digital image privacy. Image editing software’s have prepared
it incredibly simple to make changes to an image's content without leaving
any visible evidence for images in general and medical images in particular.
In this paper, the COVID-19 digital x-rays forgery classification model
utilizing deep learning will be introduced. The proposed system will be able
to identify and classify image forgery (copy-move and splicing) manipulation.
Alexnet, Resnet50, and Googlenet are used in this model for feature extraction
and classification, respectively. Images have been tampered with in three
classes (COVID-19, viral pneumonia, and normal). For the classification of
(Forgery or no forgery), the model achieves 0.9472 in testing accuracy. For
the classification of (Copy-move forgery, splicing forgery, and no forgery),
the model achieves 0.8066 in testing accuracy. Moreover, the model achieves
0.796 and 0.8382 for 6 classes and 9 classes problems respectively.
Performance indicators like Recall, Precision, and F1 Score supported the
achieved results and proved that the proposed system is efficient for detecting
the manipulation in images.