Scholarlink

The 3D extension of High Efficiency Video Coding (3D-HEVC) compresses both texture sequences and associated depth information. To optimize the coding efficiency of depth data, the standard integrates advanced inter-prediction schemes and a quadtree-based Coding Unit (CU) partitioning mechanism, denoted as depth levels. However, these enhancements considerably increase the encoder’s computational burden. This paper proposes a fast depth-level selection strategy that exploits CU homogeneity to accelerate inter-coding. At each level, we derive Average Local Variance (ALV) features and employ them in a binary machine learning model, which establishes adaptive thresholds for partitioning. The derived decision rules enable early CU termination, thereby reducing encoding complexity. Experimental evaluation confirms that the proposed algorithm significantly decreases execution time while preserving rate-distortion performance with negligible degradation.

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

Boosted Machine Learning for Fast CU Split Decision in 3D-HEVC Depth Map Inter-coding

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Othman Hdioued,
  • Siham Bakkouri,
  • Abderrahmane Elyousfi

摘要

The 3D extension of High Efficiency Video Coding (3D-HEVC) compresses both texture sequences and associated depth information. To optimize the coding efficiency of depth data, the standard integrates advanced inter-prediction schemes and a quadtree-based Coding Unit (CU) partitioning mechanism, denoted as depth levels. However, these enhancements considerably increase the encoder’s computational burden. This paper proposes a fast depth-level selection strategy that exploits CU homogeneity to accelerate inter-coding. At each level, we derive Average Local Variance (ALV) features and employ them in a binary machine learning model, which establishes adaptive thresholds for partitioning. The derived decision rules enable early CU termination, thereby reducing encoding complexity. Experimental evaluation confirms that the proposed algorithm significantly decreases execution time while preserving rate-distortion performance with negligible degradation.