Image tampering, being widely facilitated and proliferated by today's digital techniques, is increasingly causing problems concerning the authenticity of digital images. As one of the most favorable compressed media, JPEG image can be easily tampered without leaving any visible clues. JPEG-based forensics, including the detection of double compression, interpolation, rotation, etc, has been actively performed. However, the detection of misaligned cropping and recompression, with the same quantization matrix that was once used to encode original JPEG images, has not been effectively expressed or ignored to some extent. Aiming to detect such manipulations for forensics purpose, in this paper, we propose an approach based on block artifacts caused by the manipulation with JPEG compression. Specifically, we propose a shift-recompression based detection method to identify the inconsistency of the block artifacts in doctored JPEG images. The learning classifiers are applied for classification. Experimental results show that our approach is very promising to detect misaligned cropping and recompression with the same quantization matrix and greatly improves the existing methods. Our detection method is also very effective to detect relevant copy-paste and composite forgery in JPEG images.
(Publisher abstract provided.)