Improved Image Splicing Forgery Localization with First Digits and Markov Model Features

Forgery of image can be executed easily by using advanced image editing applications and the recognition of such forged images is difficult for human eyes. Image splicing is a popular way of image tampering. In this work, the goal is to recognize whether a given image is forged or not as well as localize the forged part of image. Since images of JPEG format are highly popular and broadly used, we use JPEG images as input and detect splicing as well as its localization, i.e. determining location of splicing in the image. Benford's law and Markov chain model features are extracted using first digits of DCT coefficients. These features are given to Support Vector Machine classifier for training the model at various quality factors of compression, considering two scenarios of Single Compressed Patch as well as Double Compressed Patch. Compared to state-of-the-art techniques, the experimental results of proposed work provide improved forgery detection and localization.