A Novel Approach to Image Forgery Detection Techniques in Real World Applications

Abstract

Image integrity is threatened because of the usage of modern techniques in order to manipulate the images to gain personal or monetary benefits. Image forgery has been adversely affecting the fields concerned with the usage of image as a prime source of data such as medicine and healthcare, social media, journalism and newspapers, criminal investigation, art and paintings, deep fake industry. Passive forgery is generally carried out to a greater extent in order to tamper and circulate the manipulated images. A novel design is presented which uses several Convolutional Neural Network Architectures including EfficientNetB0, VGG-16, and VGG-19 to detect copymove forging. It was trained and verified on MICC F2000 dataset and tested on MICC220 and CoMoFoD. After comparative analysis of these architectures it was found that EfficientNetB0 has the best accuracy of above 98%.