Evaluation of Image Forensic Cues for Fusion

Image forensic techniques have a sensitive role in determining authenticity of digital images. Multiple tampering operations and varied image content has led to high false positive rate of these methods when applied to real world forgeries. A detection method based on fusion of multiple cues is desired to improve reliability and accuracy in image forensics. Hence, there is an insistent need of testing the passive blind methods on standard database so that plausibility of cues for incorporating in fusion based framework can be assessed. The current paper evaluates the performance of image forensic tools that can detect cut-paste forgery by detecting inconsistency in noise. Fusion with an additional cue for detecting copy-move forgery based on intensity gradient (edge directions) has shown improvement in detection accuracy. The paper subsequently proposes an architecture of a fusion based tamper detection system based on multiple cues, which are carefully chosen to address the most common tampering traces.