Blind Median Filtering Detection Based on Histogram Features

Recently, the median filtering (MF) detector has attracted much interest as a forensic tool to identify image editing process. In this paper, we propose a novel method for the blind detection of MF in digital images based on the histogram features. As histograms are fundamental resources and can present most image information, we propose to directly utilize them by taking several highest histogram bins of the residual images as features to carry out classification. To this end, multi-scaled rotation and symmetry invariant patterns are introduced as convolution kernels for various residual images calculation and histograms generation. The effectiveness of the proposed method is verified by extensive experiments on a large image database, and the experimental results demonstrate that, with only 21 features, the proposed method outperforms some state-of-the-art works.