Image Splicing Forgery Detection Using DCT Coefficients with Multi-Scale LBP

Abstract

Image forensics is an active research area due to the large number of shared images online. These images can be easily manipulated with advanced image editing tools and the changes cannot be captured easily by bare human eyes. In this paper, a novel model is proposed based on features extracted from DCT coefficients and Multi-Scale LBP image transform to blindly detect image splicing, where two or more images are combined into one. The experiments were performed on two publicly available datasets CASIA v.1.0 and v2.0. Using k-fold cross validation, several performance measures were computed and compared with other state-of-the-art techniques. The proposed technique has demonstrated improved performance with more than 97.3% accuracy and 0.99 area under the ROC curve.