Fast CU Partitioning Algorithm for VVC Based on CNN and FSVM

• Published in: IEEE access 2024, Vol.12, p.52150-52160
• Main Authors: Song, Wenjun, Li, Congxian, Zhang, Qiuwen
• Format: Article
• Language: English
• Subjects: Algorithms; Classification algorithms; CNN; Coding; Coding standards; Complexity; Computational complexity; Convolutional neural networks; Encoding; Feature extraction; FSVM; New technology; Partitioning; Partitioning algorithms; Prediction algorithms; VVC
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3387059

The new coding standard Versatile Video Coding (VVC) introduces additional encoding techniques based on the existing video coding standard, such as the difference in block partition structures. While these new technologies bring about enhanced encoding performance, they also result in a significant increase in encoding time complexity. In the paper, we propose a decision algorithm to partition fast CUs, which is based on CNN networks and FSVM. Initially, the algorithm utilizes depth information formed by combining inter-frame correlations as input to our trained CNN model, predicting the optimal depth for encoding.Following that, the second algorithm based on FSVM is introduced. The F-Score method is employed to extract appropriate features for constructing FSVM. Within the predicted depth range from the initial algorithm, the CU partitioning mode is predicted, leading to an additional reduction in encoding complexity. Our experimental results demonstrate that the proposed algorithm can save 53.55% of encoding time, with a marginal increase of only 1.47% in BDBR.It achieves a favorable balance between video quality and encoding speed.