.png){kind=icon}
.png){kind=icon}
Return
Multiple classifier-based fast coding unit partition for intra coding in future video coding
.png){kind=icon}
PRE
AI
DOI：10.1016/j.image.2019.06.014.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Future Video Coding (FVC) significantly improves the compression efficiency over the preceding High Efficiency Video Coding (HEVC) standard but at the cost of extremely huge computational complexity. The flexible quad-tree plus binary tree (QTBT) block partitioning structure in FVC is largely responsible for the high computational complexity. In order to address the issue of huge computational complexity of QTBT, we propose a multiple classifier-based fast QTBT partitioning algorithm for FVC intra coding. The proposed multiple classifier-based QTBT algorithm contains three stages including horizon binary-tree decision model (HBTDM), vertical binary-tree decision model (VBTDM), and quad-tree decision model (QTDM). Consequently, the computational complexity of FVC intra coding can be drastically reduced by replacing the brute-force search with HBTDM, VBTDM, and QTDM to decide the optimal QTBT partitioning. To achieve high prediction accuracy, the descriptive features of intra QTBT partitioning decision are extracted to train corresponding classifiers. An efficient optimal parameter selection method of training classifiers is introduced to balance the computational complexity and rate distortion (RD) performance. Experimental results show that in comparison with the original FVC reference software, the proposed overall algorithm can reduce 64.54% intra coding time with negligible degradation of RD performance. Meanwhile, the proposed algorithm outperforms four state-of-the-art algorithms in terms of computational complexity reduction.
Keywords:
Quad-tree plus binary tree
Multiple classifier
Support vector machine
Block partitioning
Intra coding
Future video coding
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
S
IF:
2.7
Papers:
2.8K
Citations:
4.2K