Forensic analysis of full-frame linearly filtered JPEG images

The characteristic artifacts left in an image by JPEG compression are often exploited to gather information about the processing history of the content. However, linear image filtering, often applied as postprocessing to the entire image (full-frame) for enhancement, may alter these forensically significant features, thus complicating the application of the related forensics techniques. In this paper, we study the combination of JPEG compression and full-frame linear filtering, analyzing their impact on the Discrete Cosine Transform (DCT) statistical properties of the image. We derive an accurate mathematical framework that allows to fully characterize the probabilistic distributions of the DCT coefficients of the quantized and filtered image. We then exploit this knowledge to estimate the applied filter. Experimental results show the effectiveness of the proposed method.