Scholarlink

<p>The application of new coding standards brings a wider range of applications as well as new video forensic challenges. In order to solve the transcoding detection problem under the new video standard H.266/Versatile Video Coding (H.266/VVC), we propose a new algorithm based on Coding Unit (CU) block type classification and Motion Vector Map (MVM) statistics to detect VVC transcoded videos. We first analyze the generation of HEVC/VVC transcoded video traces and define MVM graphs for constructing features. By calculating the average frequency of different CU partition types for intra-frame prediction frames and inter-frame prediction frames, they are connected and sent as distinguishing features to a support vector machine for classification. Experimental results show that the method has high recognition accuracy under the new coding standard. Specifically, the algorithm constructs a 51-dimensional feature vector by combining intra-frame and inter-frame features, capturing transcoding traces such as changes in CU partitioning and motion vector distributions. Extensive experiments on 1080p and 4K video datasets, across various bitrate combinations, quantization parameters, and Group of Pictures (GOP) structures, demonstrate an average detection accuracy of 96.81% on 1080p datasets, outperforming existing methods. The robustness of the proposed approach across diverse encoding configurations highlights its effectiveness for forensic analysis under the VVC standard.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detection of transcoding from HEVC to VVC based on CU types and Motion Vector Map

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Jiajian Lin,
  • Jianlin Zhang,
  • Shan Bian,
  • Chuntao Wang

摘要

The application of new coding standards brings a wider range of applications as well as new video forensic challenges. In order to solve the transcoding detection problem under the new video standard H.266/Versatile Video Coding (H.266/VVC), we propose a new algorithm based on Coding Unit (CU) block type classification and Motion Vector Map (MVM) statistics to detect VVC transcoded videos. We first analyze the generation of HEVC/VVC transcoded video traces and define MVM graphs for constructing features. By calculating the average frequency of different CU partition types for intra-frame prediction frames and inter-frame prediction frames, they are connected and sent as distinguishing features to a support vector machine for classification. Experimental results show that the method has high recognition accuracy under the new coding standard. Specifically, the algorithm constructs a 51-dimensional feature vector by combining intra-frame and inter-frame features, capturing transcoding traces such as changes in CU partitioning and motion vector distributions. Extensive experiments on 1080p and 4K video datasets, across various bitrate combinations, quantization parameters, and Group of Pictures (GOP) structures, demonstrate an average detection accuracy of 96.81% on 1080p datasets, outperforming existing methods. The robustness of the proposed approach across diverse encoding configurations highlights its effectiveness for forensic analysis under the VVC standard.