JPEG Implementation Forensics Based on Eigen-Algorithms

Due to the widespread diffusion of JPEG compression standard, image forensic researchers have developed a series of techniques to reverse engineer specific JPEG-based information (e.g., the used quantization matrix, presence of double JPEG compression, JPEG grid alignment, etc.). In this paper, we focus on the detection of traces left on images by the use of different JPEG implementations (e.g., characteristic of proprietary software suites). Specifically, given a JPEG image under analysis, we propose a compact descriptor that enables to distinguish which JPEG implementation has been used. This is done considering the challenging scenario in which the same quantization matrix is used. Results show that it is possible to distinguish between two popular JPEG implementations, as well as to adapt the proposed methodology to additional JPEG-based forensic problems.