A Robust Seam Carving Forgery Detection Approach by Three-Element Joint Density of Difference Matrix

摘要

Seam carving is a popular content-aware image retargeting technique. However, it can also be used for malicious purposes such as object removal. In this paper, a robust blind forensics approach is proposed for seam-carved forgery detection. Since insignificant pixels along seams are removed for image resizing, the spatial neighborhood relations among pixels will be significantly changed, especially in smooth regions. Thus, joint density is exploited to model the change of spatially adjacent pixels' distribution caused by seam carving, even in the case of low scaling ratios. Specifically, three-element joint density of difference matrix is computed to form general forensics features (GTJD). The GTJD features are combined with existing energy and noise features exacted in LBP domain for classification. Experimental results show that the proposed approach achieves better accuracies for both uncompressed images and JPEG images with different scaling ratios.