The explanation of the noun in HEVC

the explanation of the noun in HEVC

noun explanation

In order to understand the details of the HEVC encoder, it is necessary to have a clear understanding of the following nouns.

1) LCU

A coded image is usually divided into a number of basic units that can be set at their own size, called LCU (largist Coding unit). The size of the LCU can be set to a range of 16x16-64x64.

2) Slice

The slice is a sequence of CTU in a frame-by-raster scan sequence, one frame can be composed of multiple slice, each slice can be decoded independently, because the prediction of slice pixels cannot cross the slice boundary. Each slice can be divided into i/p/b slice according to the different encoding types. The main purpose of this structure is to achieve resynchronization after encountering data loss in the transmission. The maximum number of bits that can be carried per slice is usually limited, so depending on the movement of the video scene, the amount of CTU that slice contains can vary greatly.

3) Dependent Slice

Dependent Slice, whose decoding or encoding of the starting Entropy encoding Cabac context state is based on the above one slice, so it cannot complete the resynchronization after data loss, the technology can be understood as a further separation of the original Nalu data, can be suitable for more flexible packaging methods.

4) Tile

Tiles are rectangular areas within the frame that can be decoded independently, containing multiple, rectangular-arranged CTU (the number does not necessarily have to be the same), which is defined as the intention of enhancing the parallel processing performance of the codec, where multiple tiles within the same slice can share the same header information, A tile can also contain multiple slice, whether slice contains tiles, or the tile contains slice must be an integer-containing relationship, the tile diagram is as follows:

5) WPP

WPP's parallel technology is based on a line of LCU blocks, but does not completely truncate the relationship between LCU rows, as shown below, the CABAC state of the second block of Thread1 is saved, used to Thread2 the initial Cabac state, and so on parallel encoding or decoding, Therefore, there is a large dependency between rows and rows. Generally, this method is more compressible than tiles.

6) CU

A LCU is usually divided into several Cu (Coding Unit), and a LCU is represented by the CU_SPLIT flag to indicate whether a downward split is required. If the Cu_split FLAG=1,LCU is split with a quadtree, it gets four cu of the same size (the length width is half of the pre-split). Split each down once, recording the depth of the split depth plus 1. A CU consists of a luminance pixel block and an additional CB and a CR color pixel block (the length of the chroma pixel is half the brightness).

7) PU

PU (prediction Unit), the predictive mode for each CU may be within or between frames.

If the prediction mode is within the frame, the size of the PU is equal to CU in addition to the minimum size mode. If the minimum size mode is used, the flag bit indicates whether the CU is divided into four pu blocks. This minimizes the ability to support intra-frame prediction blocks of 4*4 size (at this point the Chroma component Pu is also 4*4).

If the prediction mode is inter-frame prediction, the luminance and chroma cu may be split into 1/2/4 pu. Only CU is a minimum size mode when it can be split into 4 pu blocks (and the minimum size is not 8*8). When CU is divided into four Pu, each pu occupies CU 1/4;cu divided into two Pu, can choose one of 6 kinds of segmentation mode, which contains the average split mode amp (only for 16*16 and above the brightness block). The prediction between each frame Pu contains one or two motion vectors and a reference frame index.

8) TU

On the basis of the CU can be the TU block recursive division, TU allowable size range and 32*32-4*4, but the maximum can not exceed the current TU block is the size of the CU block, so usually the range of Tu block is the value of cu-size-4, when the frame pu is n*n, tu must be divided down a layer When the inter-frame mode is non-2n*2n, and the tu depth is 0, it must be divided down one layer at a time.

9) Z-Scan

There is also a grating scanning sequence in HEVC, but it is not convenient for the CU to be addressed in the way that the CU is recursively divided, if the order of the raster scan is still used. Therefore, HEVC defines the z-scan order.