Transfer Learning Approach for Splicing and Copy-Move Image Tampering Detection

Image authentication before using in any security critical applications has become necessary as the image editing tools are increasing and are handy to use in today's world. Images could be tampered in different ways, but a universal method is required to detect it. Deep learning has gained its importance because of its promising performance in many applications. In this paper a new framework for image tampering detection using Error Level Analysis (ELA) and Convolutional Neural Network (CNN) with transfer learning approach is proposed. In this method, the images are pre-processed using ELA to highlight the tampered region and are used to fine tune the entire model. Six different pre-trained models are used in the proposed framework to compare the performance in classifying the tampered and authentic images. The complexity and processing time of the proposed method is low with respect to most of the existing methods as the images are not divided into patches. The performance of the model obtained is also considerably good with an accuracy of 97.58% with Residual Network 50(ResNet50).