Detection of Median Filtering Manipulation in Compressed Images

Median filtering, as a nonlinear content preserving process that is often employed for smoothing and denoising images, has received significant attention from the forensics and documentation research community. In this paper, a detection scheme for median filtering of compressed images is proposed, based on singular value decomposition of the process matrix. In the proposed method, the process matrix is obtained by linear estimation of the decoding process, implementation of median filtering and recompression of the image. The projections of input data over eigenspaces of this process matrix is then used as features of the image. A small number of such features are utilized to classify the image as either original or processed, thus leading to a fast and effective detection scheme for median filtering. The experimental evaluations show that the proposed scheme outperforms the existing methods, particularly over highly compressed images, and its detection error is 2% to 5% lower in comparison to the errors introduced by other detection schemes. The singular value decomposition of the process matrix introduced in this paper, may also be used in detection of other cases of image content manipulation.