LAUSR

Median filtering is one of the most common operations for image smoothing and retouching, and it is often used as a post-processing by forgers to alleviate the traces of image tampering. Hence, if the traces of median filtering can be found in an image, this image is highly suspected. In this paper, we proposed an adversarial network for median filtering detection in RGB images. Our detection framework can be divided into three parts. To overcome the previous limitations of median filtering detection in gray-scale images, we first extract the dark channel residual in RGB images for suppressing the interference of content. Second, we merge several dark channel residual together by multi-scale fusion to better characterize the statistic traces left by different filter sizes. Third, we explore a generative adversarial network to improve the robustness and enhance the statistical difference between original images and median filtered images. Our method is extensively evaluated in several publicly available data sets. The experimental results present an obvious improvement compared with other competitors. Particularly, the proposed framework obtains better performances in the case of the small blocks with JPEG compression.