Passive forensics for image splicing based on PCA noise estimation

In this paper, based on the fact that images from different sources tend to have different noise levels, an effective method for detecting image splicing is proposed by detecting the inconsistency of local noise levels. The local noise level is estimated by a novel blind noise estimation method on the basis of the principal component analysis (PCA). We describe three detection schemes to get the local noise map. Then, the observed image is segmented into various regions with homogenous noise levels according to different local noise variances. Experimental results demonstrate that the proposed method can effectively localize the tampered region with higher detection accuracy compared to other methods in literatures.