Image recapture detection through residual-based local descriptors and machine learning

Abstract

At present, the tamper evidence would be invalid in recaptured image in terms of most of the digital image forensics, so the authenticity of the image detection is a security threat. Since dense local descriptors and machine learning have been successfully applied in steganalysis and forgery detection, we propose a new image recapture detection method based on these two techniques. The local descriptors were recently proposed in the field steganalysis, and some descriptors are selected by greedy strategy in the experiments. Support vector machine and ensemble classifier are utilized as the classifier in the proposed method. The experimental results show that the proposed method achieves a good performance rate that exceeds 99.61% of recaptured images and 96.40% for single captured images on the open source database.