Multiple JPEG detection using convolutional neural networks in the DCT domain

Abstract

The recent rapid growth of social network services (SNSs) has changed the way in which images are shared. Since images undergo JPEG encoding and decoding many times through the repeated uploads and downloads, multiple JPEG compression detection is becoming more critical in the digital image forensic field. Existing methods are based on the statistical characteristics of images and do not utilize recent advances in deep learning. Moreover, the traces of JPEG compression are small. The more compression occurs, the harder it is for the classifier to detect

it is difficult to identify the number of compressions by adopting well-defined convolutional neural networks (CNNs) on computer vision field. In this paper, we propose a novel CNN for multiple JPEG compression detection on the 2D discrete cosine transform (DCT) histogram. The proposed method achieves higher performance than state-of-the-art works and shows considerable experimental results for a practical scenario using SNS platforms.