摘要: Image compression evidence has been utilized as an important forensic feature to justify image authenticity. However, some recent studies showed that the of block transform-based coding, e.g., JPEG and JPEG2000, can be effectively erased by adding designed dither noise in transform domain. In this paper, we demonstrate it is also feasible hide lossy predictive a class paradigm widely employed critical scenarios. To tackle challenging issue error propagation inherent design prediction-direction preserving strategy, allowing us add prediction (PE) domain, while minimizing incurred distortion. Extensive experimental results are provided verify effectiveness proposed anti-forensic algorithm for coding.