Detection of H.266/VVC video transcoding based on refined block partition and filtering modes statistics in coding domain

Video transcoding has posed significant challenges to the authenticity and integrity of digital videos, which makes the detection of transcoded videos one of the most important video forensics techniques. However, the detection of transcoded H.266/Versatile Video Coding (H.266/VVC) videos is rarely reported in the literature. To tackle this issue, we first analyze the generation of VVC transcoding traces. It is found that the variation of the block partition and filtering modes in video frames can be used as the clue to expose VVC transcoding. Then a series of refined coding technologies in VVC, such as Multi-type Tree (QTMT) partition and new in-loop filtering, are analyzed to facilitate feature selection. On this basis, the proportions of Coding Units (CUs) with different sizes in intra-coded frames, filtering modes in Deblocking Filtering (DBF), positive and negative offsets in Sample Adaptive Offset Filtering (SAOF), and blocks with different categories in Adaptive Loop Filtering (ALF) are used for feature construction. These feature sets are further concatenated into a 61-dimensional feature vector for supervised learning and video classification. Extensive experiments are conducted to evaluate the performance. The results demonstrate the effectiveness and robustness of our method in transcoding detection.