Copy-Move Forgery Detection using Residuals and Convolutional Neural Network Framework: A Novel Approach

With the sudden advancement in digital image processing, there has been a huge upsurge in the creation of doctored or tampered images with the successful aid of softwares like GNU Gimp and Adobe Photoshop. These manipulated images have become a serious cause of concern, especially in the news, politics and the entertainment sector. Therefore, there is an alarming requirement for a robust image tampering detection system which can distinguish between authentic and tampered images. Common image tampering techniques include copy-move forgery, seam carving, splicing and re-compress. Amongst these techniques, copy-move forgery detection (CMFD) and splicing are dominating the research field due to their complexity stratum and difficulty in detection. In this work, we focus on proposing an efficient splicing detection and CMFD pipeline architecture that focuses on detecting the traces left by various post-processing operations of Splicing and copy-move forgery that are JPEG Compression, noise adding, blurring, contrast adjustment, etc. We use second difference of median filter (SDMFR) on the image as one of the residual and the Laplacian filter residual (LFR) together to suppress image content and focus only on the traces of the tampering operations. The proposed method achieves higher accuracy of 95.97% on the CoMoFoD dataset and 94.26% on the BOSSBase dataset.