Image composition detection using object-based color consistency
Gang Cao, Yao Zhao, Rongrong Ni · 2008
An efficient approach for the blind image composite detection is proposed in this paper. Image composite is commonly used when an image forgery occurs. A forgery detector is devised to expose this kind of tampering through checking consistency of the color perception between object area and background area. Local color statistic and object-based color temperature metric are presented to assess color realism over the whole image. Low realism score is considered as the evidence of tampering. Tests on a refined composite image dataset demonstrate a promising performance with 70% precision under lower false positive possibility.