One-Class HEVC Double Compression Detection with Same Coding Parameters
Yulin Zhao, Xiangling Ding · 2024
High Efficiency Video Coding (HEVC) double compression detection with the same coding parameters can be regarded as one principal procedure to analyze the integrity of HEVC-coded videos. Therefore, a one-class classification (OCC)-based hybrid heterogeneous network with a shallow convolutional neural network (CNN) and a six-node graph neural network (GNN), which only needs the pristine videos, is proposed in this paper. Concretely, the shallow CNN learns the subtle fluctuation of pixel values due to double compression, while the six-node GNN is developed to represent the local-global relationship of various divided patches and the number and position of zero-value pixels in a high-frequency component of the motion-aligned residual. Finally, the output vectors are fused into a specially designed semi-supervised Mahalanobis distance-based OCC to obtain the detection results. The experimental result demonstrates that the proposed method, which only learns features from single compressed videos, outperforms the existing state-of-the-art detection methods and other more complex OCC methods.