JPEG Image Security by Block Size Estimation and Quality Factor Classification

Passive methods in compressed images forensics are more preferable than active ones like watermarking. In passive methods, the deviation of natural characteristics of the compressed image plays a key role in performance. In this paper, based on the blockiness effect and its corresponding discontinuity imposed on block edges, we propose a universal normalized feature for passive analyses. Using this feature, we devise a high-precision straightforward procedure for JPEG block size extraction, and a machine learning approach for JPEG quality factor estimation. We also show that by extracting the messiness parameter of the image and designing a two-stage classifier, it is possible to boost the classification performance. Our simulation results validate the accuracy and efficiency of the proposed scheme.