A Robust Copy Move Forgery Classification Using End to End Convolution Neural Network

Abstract

in this digital world it is not surprising to do manipulation with digital images. With advantage of such technologies it has become very easy to misguide the observer about the reality appearing in the images. The objective behind such manipulation for fun and entertainment is acceptable but when such things are applied on sensitive information such as evidences used in judiciary system, to prove certain claim, using such manipulated images on social media becomes dangerous. Although many types of forgeries that could be performed with images such as copying certain part of image then pasting it in same image somewhere else in document with such precision that it appears normal to observer called copy move forgery. Other forgeries include splicing of images, image morphing, retouching etc. Two different categories of approaches are being used till recently for this problem of copy-move forgery detection. These are block based and Keypoint based approach wherein block based is computationally intensive and suffers from many disadvantages and other is based interest points or high textured areas whose features vectors are formed for comparison to find the duplicated regions. In this paper, a deep neural network based approach has been proposed with promising results that can classify images based whether any copy move forgery has been there in the images. The proposed work aims to classify all the images having copy move forgery with presence of scaling, rotation, different compression level. A new CNN model has been researched for this problem to obtain the accuracy of around 93-95 percent for different datasets alone as well on the combination of two or more datasets. © 2020 IEEE.