Fast coding unit partitioning algorithms for versatile video coding intra coding

Jiunn Tsair Fang, Bang Hao Liu, Pao Chi Chang

Research output: Contribution to journalArticlepeer-review


Versatile video coding (VVC) is the newest video compression standard. It adopts quadtree with nested multi-type tree (QT-MTT) to encode square or rectangular coding units (CUs). The QT-MTT coding structure is more flexible for encoding video texture, but it is also accompanied by many time-consuming algorithms. So, this work proposes fast algorithms to determine horizontal or vertical split for binary or ternary partition of a 32 × 32 CU in the VVC intra coding to replace the rate-distortion optimization (RDO) process, which is time-consuming. The proposed fast algorithms are actually a two-step algorithm, including feature analysis method and deep learning method. The feature analysis method is based on variances of pixels, and the deep learning method applies the convolution neural networks (CNNs) for classification. Experimental results show that the proposed method can reduce encoding time by 28.94% on average but increase Bjontegaard delta bit rate (BDBR) by about 0.83%.

Original languageEnglish
Article number103542
JournalJournal of Visual Communication and Image Representation
StatePublished - Aug 2022


  • Coding unit
  • Convolutional neural network
  • Intra coding
  • Quadtree with nested multi-type tree
  • Versatile video coding


Dive into the research topics of 'Fast coding unit partitioning algorithms for versatile video coding intra coding'. Together they form a unique fingerprint.

Cite this