The default H.266 CTU size of 128 * 128; CTU CU is divided into the size of 128 * 4 * 128-4; unless CU is further divided into a size greater than the maximum allowed length transform, or not in the prediction and transform CU PU and into the TU;
Reference: https://blog.csdn.net/huster1446/article/details/87907480
The image is divided into a CTUs
VVC in a video frame is divided into a number of CTU. CTU concepts and HEVC same. For a three-channel image frame, comprising a CTU N × N luminance samples and two corresponding blocks of chroma samples. That is the same size of each of CTU. As shown below:
VVC luminance component CTU maximum size is 128 × 128 (HEVC is 64 × 64). Note that the VVC maximum brightness transform block size is 64 × 64.
CTU division
In HEVC, CTU divided by a quad-tree structure into a plurality of sub-CU, also referred to as a coding tree (coding tree), it is done to be able to adjust for the different characteristics of spatial encoding (intra or inter) . In sub-step decision CU is intra-frame coding (using the spatial correlation) or inter-coding (using a temporal correlation).
Each CU may be further sub-divided into one, two or four PU, depending on the PU partition type. In a PU sample inside the same prediction mode, predictions related to the unit of PU is also transmitted to the decoding side. After the prediction of the PU and to obtain residual information, a sub-CU may be further divided into a number of the TU, which is divided by a quadtree completed. HEVC is a key feature is divided into a variety of concepts: CU, PU and TU.
Without CU, PU and TU concepts in the VVC, uniform application of the concept of a coding unit (coding unit, CU) is. Unless the current coding block size is larger than the maximum length supported by the transformed VVC (i.e. in this case be divided into CU TU). While using multiple embedded type tree structure quadtree dividing CTU, support more shape CU (except there is a rectangular square).
CTU will first quadtree division, then further divided using a multi-type tree (binary, trigeminal). Multi-type tree structure contains four kinds of division manner (see below): a vertical binary division (SPLIT_BT_VER), the level of the binary division (SPLIT_BT_HOR), the vertical division of the trigeminal (SPLIT_TT_VER), trigeminal horizontal division (SPLIT_TT_HOR). Child node is called a division CU. Unless a size greater than the maximum transformation CU length allowed, otherwise not be further divided into CU PU and TU in predicting and transformation. That most cases, CU, PU, TU sizes are the same.
The following figure illustrates the quad-tree structure dividing mechanism embedded multi-type tree. CTU is considered as the root node of the quadtree and divide. Each child node of the quadtree can be further purified by multiple types of trees (MTT) is divided. First, to indicate whether it will be divided by MTT mark (mtt_split_cu_flag). If further divided, the division is represented by the direction mark (mtt_split_cu_vertical_flag), and is represented by a symbol (mtt_split_cu_binary_flag) is a binary divide or trigeminal division. Both markers can be determined partition type of MTT. Dividing mechanism schematic diagram as follows:
MTT obtained from the flag pattern is divided as follows:
The figure is divided into CTU CU of FIG. Bold border blocks represented by the quadtree division obtained, not bold represented by the MTT partitioning obtained.
MTT embedded quad CU divided such that the shape can adapt to the content of the image. The maximum size of the luminance component CU CTU same size and can be a minimum of 4 × 4. If 4: 2: YUV sequence 0, the CB chroma component of at most 64 × 64, a minimum of 2 × 2.
In the VVC, a maximum luminance component supported transform size of 64 × 64, the maximum supported chrominance component transform size of 32 × 32. If the size of a CB exceeds the limit, then the level will automatically CB or (and) vertical division to meet transform size corresponding direction.
The following is the syntax elements related to SPS MTT embedded quadtree division related to:
CTU size: Size quad root node
MinQTSize: quadtree child node minimum size allowed
MaxBtSize: the root node of the binary tree as the maximum size allowed
MaxTtSize: trigeminal root node maximum size allowed
MaxMttDepth: from a quadtree the maximum depth of a child node can be divided MTT
MinBtSize: binary tree child node minimum size allowed
MinTtSize: trigeminal child node tree minimum size allowed
Next illustrates quadtree structure is embedded MTT:
for 4: 2: 0 the YUV sequence, CTU default size is 128 × 128 (i.e., a luminance of 128 × 128 64 × 64 CTB and two chrominance CTB), MinQTSize is 16 × 16, MaxBtSize is 64 × 64 (here in the original document is 128 × 128, I think it should be wrong), MaxTtSize is 64 × 64, MinBtSize and MinTtSize for the 4 × 4, MaxMttDepth is 4.
CTU is first divided by four forks, the size range child node of the quadtree is from 16 × 16 (MinQTSize) to 128 × 128 (CTU size). If the child node of the quadtree size of 128 × 128, then it will not be performed and trigeminal binary division (because it exceeds the MaxBtSize and MaxTtSize, i.e., 64 × 64). If the child node of the quadtree size than 128 × 128, then the quadtree nodes may be divided into further sub MTT, in this case, the child node of the quadtree MTT is the root node (mttDepth 0). When the depth of the dividing MTT arrival MaxMttDepth (i.e., 4) that no further division. When MTT MinBtSize node and a width equal to or less when 2 * MinTtSize (individual should be appreciated that these two conditions are used to further restricting binary division and the division of the trigeminal), will no longer be horizontal division. Similarly, when the child node MTT height equal to and less than or equal to 2 * MinBtSize when MinTtSize, will not be divided vertically.
Meanwhile, in order on the hardware decoder VVC 64 × 64 block of luminance and chrominance blocks of 32 × 32 design enables the pipeline is prohibited when the luminance block is wider or taller than 64 ternary tree (TT) is divided. When the width or height greater than 32 chroma block is prohibited TT division. As shown below:
In VTM3, the luminance and chrominance components of the block division structure can be different, and now (VTM4) is, for P slices and B slices, the same luminance and chrominance CTU of CTB CTB division must have the same structure, I slices luminance and chrominance components are divided can have a separate structure.
When the divided structure luminance and chrominance components are different, the luminance and chrominance components using two CTB coding tree structure divided into CU. That is in CU I slice may contain only a luminance component of the CB, CB or comprises only two chrominance components. And P slice and B slice is always CU comprises a luminance component and two chrominance components CB CB (the terms for the three color components of a video sequence).