Improved JPEG anti-forensics with better image visual quality and forensic undetectability

- A JPEG anti-forensic scheme is proposed to fool the existing forensic detectors.
- The proposed scheme hides the JPEG artifacts in both spatial and DCT domain.
- The proposed denoising algorithms improve the image visual quality.
- Better forensic undetectability is achieved by the improved deblocking operation.
- A better tradeoff is achieved between image quality and forensic undetectability.

There is an immediate need to validate the authenticity of digital images due to the availability of powerful image processing tools that can easily manipulate the digital image information without leaving any traces. The digital image forensics most often employs the tampering detectors based on JPEG compression. Therefore, to evaluate the competency of the JPEG forensic detectors, an anti-forensic technique is required. In this paper, two improved JPEG anti-forensic techniques are proposed to remove the blocking artifacts left by the JPEG compression in both spatial and DCT domain. In the proposed framework, the grainy noise left by the perceptual histogram smoothing in DCT domain can be reduced significantly by applying the proposed de-noising operation. Two types of denoising algorithms are proposed, one is based on the constrained minimization problem of total variation of energy and other on the normalized weighted function. Subsequently, an improved TV based deblocking operation is proposed to eliminate the blocking artifacts in the spatial domain. Then, a decalibration operation is applied to bring the processed image statistics back to its standard position. The experimental results show that the proposed anti-forensic approaches outperform the existing state-of-the-art techniques in achieving enhanced tradeoff between image visual quality and forensic undetectability, but with high computational cost.