Hiding Traces of Camera Anonymization by Poisson Blending

Sensor noise caused by photo response non-uniformity (PRNU) has been widely accepted as a reliable fingerprint for source camera identification (SCI). An interesting research topic in this area concerns the repudiability of PRNU-based SCI, which includes methods of removing or synthesizing the fingerprint on which forensic methods rely. Removing the PRNU fingerprint from a given image, also known as camera anonymization, is important for privacy protection and anti-tracking. However, camera anonymization sometimes introduces annoying visual artifacts in the resultant image. In this work, Poisson blending is used to hide the traces left by camera anonymization. Theoretical analysis and experimental results show that the proposed method can suppress the visual artifacts caused by camera anonymization effectively while maintaining the anti-forensic effectiveness.