Effect of JPEG compression on Sensor-based Image Forensics

Abstract

For the forensic task of camera identification and manipulation localization, the most promising and widely used algorithms are the ones based on Photo-Response Non-Uniformity (PRNU) or the camera sensor noise. However, the performance of camera identification and manipulation localization is strongly affected by JPEG compression. Typically, for images with lower quality factors, the accuracy of camera identification as well as manipulation localization reduces considerably. In this article, we present the results of a large scale demonstration on how JPEG compression has an influence on the performance of camera identification and manipulation localization. Our results on three popular datasets indicate that camera identification works even for a quality factor as low as thirty. Also we found that false positives increase significantly for PRNU-based manipulation localization.