Robust content authentication of gray and color images using lbp-dct markov-based features

Abstract

This paper presents a robust method for passive content authentication of gray and color images. The idea is to capture local and global artifacts resulting from the image manipulation through combining intra-block Markov features in both LBP and DCT domains. An optimized support-vector machine with radial-basis kernel is then trained to classify images as being tampered or authentic. We intensively investigate the authentication capabilities of the proposed method for separate color channels and for various combinations of them. The proposed method, without and withfeature-level fusion, is evaluated on three benchmark datasets with a variety of forgery and post-processing operations. The results show that fusing Markov features from LBP and DCT modalities leads to consistent improvement in terms of detection accuracy as compared to the state-of-the-art passive methods. Furthermore, using information from all YCbCr channels help enhancing the detection rate to more than 99.7 % on CASIA TIDE v2.0 image collection.