JADE OWL: JPEG 2000 Forensics by Wavelet Offset Consistency Analysis

Abstract

In a world teeming with digital images, the credibility of visual data has become of paramount importance. While it is now simpler than ever to manipulate an image for malicious purposes such as misinformation, the tools for detecting such alterations have predominantly been developed for either uncompressed or JPEG-compressed natural images. However, medical and satellite imagery, domains where the potential for fraud is high, often use a different compression format -JPEG 2000. We present a JPEG 2000 Anomaly Detection Estimator via Offset of Wavelet Localization -the Jade Owl -, a novel method for detecting forgeries in JPEG 2000 images by analyzing the consistency of traces left by its compression. Our technique hinges on the premise that the wavelet coefficients of a JPEG 2000 image are lower when the same offset is applied during the wavelet transform than they are when the offset is different. By employing this principle locally, we're able to detect regions with significantly different offsets, indicating potential forgeries such as copy-move. An accompanying a contrario model further refines this detection to make automatic detections while controlling false positives. To evaluate the method, we've created a unique dataset of JPEG 2000 forgeries. This novel approach significantly paves the way for JPEG 2000 image forensics, introducing a sensitive and efficient tool for authenticity verification in critical sectors such as healthcare and satellite imagery.