FORENSIC METHODS FOR IMAGE AND VIDEO INTEGRITY VERIFICATION

The detection of double compression is an effective method for identifying tampered multimedia content. Tampering involves decoding, modifying, and re-encoding the original content. This re-encoding, known as double compression, leaves distinctive traces that do not exist in the original. By analyzing these traces, tampering can be detected. In recent years, research in image forensics has primarily focused on analyzing double-compressed images in the JPEG format. However, effective methods for detecting double compression in the High Efficiency Image File Format (HEIF), which has gained popularity for its ability to maintain high quality with reduced file size, remain limited. As HEIF employs H.265/HEVC encoding, traditional JPEG-based approaches cannot be directly applied. In response, we previously proposed a detection method based on coding ghost theory, although this method was restricted to cases where the quality of the first encoding was higher than that of the second. To overcome this limitation, we developed a lightweight image classifier capable of detecting double compression in HEIF images without relying on the quantization history. This model achieves higher detection accuracy than traditional methods while maintaining its lightweight characteristics. Similarly, the importance of double compression detection in video forensics has grown, particularly for the HEVC format, although research on double compression detection in video sequences with B-frames remains limited. Video tampering through double compression may cause frames to be re-encoded with different frame types from the original encoding, disrupting spatiotemporal consistency and potentially leading to anomalous behavior in intra-frame encoding elements. This effect leaves detectable traces in the encoding structure that were not present in the original video. In this study, we investigate double compression in both HEIF images and HEVC videos with B-frames, proposing a classifier based on changes in encoding information caused by re-encoding. Our method demonstrates superior detection accuracy compared to conventional techniques and proves effective in real-world scenarios.