A variational approach to JPEG anti-forensics

The objective of JPEG anti-forensics is to remove all the possible footprints left by JPEG compression. By contrary, there exist detectors that attempt to identify any telltale of the image tampering operation of JPEG compression and JPEG anti-forensic processing. This paper makes contribution on improving the undetectability of JPEG anti-forensics, with a higher visual quality of processed images. The employment of constrained total variation based minimization for deblocking successfully fools the forensic methods detecting JPEG blocking, and another advanced JPEG forensic detector. Calibration-based detector is also defeated by conducting a further feature value optimization. Experimental results show that the proposed method outperforms the state-of-the-art methods in a better trade-off between forensic undetectability and visual quality of processed images.