Performance Analysis of ELA-CNN model for Image Forgery Detection

With the growth of technology, the amount of content being shared on day-to-day basis is a figure beyond our comprehension. This data can be shared in the form of photos and videos and can be spread using social media platforms like WhatsApp and Instagram. With this also comes the threat of forged images being used to share false information. Advanced tools and software have made it easy and quick to generate such images. Due to this, researchers have been focusing on devising methods to tackle the same in a hope to assist people in finding the authenticity of questionable images. Copy-move and splicing are two forms of commonly used forgery techniques. When a region of the image is copied and pasted on the same image it is referred to as copy-image forgery whereas a region of a different image is copied and pasted on another image it is said to be spliced forgery. Our technique makes use of error level analysis (ELA) that points out the difference between the compression ratios of authentic and manipulated image and hence highlights the latter. The extracted information is then fed into a variant of CNN which classifies the image as original or forged. We also provide a comparative analysis of the proposed method with statistical and deep learning models and it will be tested on the CASIA V2, MICC-F220 and MICC-F2000 forged image datasets.