Detection of video transcoding from AVC to HEVC based on Intra Prediction Feature Maps

Abstract

As the High Efficiency Video Coding (HEVC) standard gains popularity, forgers are more inclined to transcode videos into HEVC format from the previous Advanced Video Coding (AVC) format. To verify the originality and authenticity of videos, it is crucial to propose a method for transcoded HEVC video detection. In this paper, a novel method is proposed to detect video transcoding from AVC to HEVC (AVC-HEVC). Analysis shows that the intra prediction mode is sensitive to spatial loss introduced by previous AVC encoding in the transcoding process. Thus, the intra prediction modes are extracted from the luminance and chrominance components to create Intra Prediction Feature Maps (IPFMs). Subsequently, a Dual-flow Attention-based MobileNet (DAM-Net) is introduced to learn the deep representation of AVC-HEVC transcoding artifacts. Finally, video level results are derived from the frame level analysis provided by DAM-Net. Extensive experiment results demonstrate that the performance of the proposed method outperforms the existing methods in the detection of AVC-HEVC transcoding.