On the benchmarking of ResNet forgery image model using different datasets

This paper presents the benchmarking and improve-ment of the ResNet image forgery model using three different datasets (CASIA, Columbia, and LSBU). The model is based on classification, where forgery images have been edited using cut-paste modification technique. The images are categorized to check if the algorithm can successfully identify the difference between the original and the forgery image. All images have been pre-processed with Gray-Edge detectors to obtain get better classification results. Experimental results have shown that the Gray-edge technique has improved the accuracy across all image datasets.