Source: The H series has MPEG and VCEG launched, the VPX series is launched by Google, H.265 has a higher image quality, while AV1 is more reliable and free for streaming media. The H.265's larger predictive model enables edge visualization, while VP9 implements a stricter coding rule, which seems to make the streaming media more coherent and reliable. Among them, H.265 has a compression efficiency ratio H.264 by 50%, and VP9 is slightly incomplete H.265, the H.266 has the highest encoding performance, and the AV1 is reduced by 30% by the same quality of VP9.
content:
Compression performance comparison
0. Code stream structure
Head information
2. Code frame difference
3. Block division structure
4. Inframe prediction
5. Inter-frame prediction
6. Transformation
7. Entropy encoding
8. Filter
9. Acceleration technology
10. Other technology
Be
0. Code stream structure:
H.264:
NAL layer: Network abstraction layer mainly used for network transmission. According to a certain format, the data outputted by the video coding layer is packaged and packaged, and information such as header is provided to transmit or store on a different rate network.
Video Coding Layer: Video coding layer mainly used for data coding, NAL provides a layer of protection around VCL. In H.265 / HEVC, the NAL unit is divided into Vclu (Video Coding Layer Nal Unit) and Non-Vclu based on whether or not the loading video coded data is loaded.
Head information
H265 Composition of an image sequence: VPS + SPS + PPS + SEI + A frame + Several P frame. VPS, SPS, PPS, SEI, one I frame, one P frame can be referred to as a NALU. Transfer from https://blog.csdn.net/qq_39660930/Article/details/78224185
2. Code frame difference
H.265 still uses a mixed codec, the codec structure is basically consistent with H.264
H.265 Classic Frame:
Be
3. Block division structure:
H.264 is 16x16 (the size of the sub-block can be 8x16, 16x8, 8x8, 4x8, 8x4, 4x4 very flexible) macroblock,
VP9 can be sampled in a block of 64 × 32 or 4 × 8, supporting 64 × 64, supporting the frame to be divided into a region having a specific similarity; compared to H.265, VP9 supports horizontal or vertical subdivision
H.265 is a recursive structure of Cu (CodingUnit), PU (PREDictionUnit) and TU (TransformUnit) (predictive block brightness 64x64-8x8, chromaticity 32x32-4x4, transmends 32x32->4x4), At the same time, H.265 increases asymmetric division mode; the specific segmentation process is marked by two variables: split depth (depth) and split marker (split_flag), H.265 / HEVC standard breaks through the previous criteria for predictive blocks, transform Limit of a large relationship between blocks. Since its PU and TU are directly obtained by CU, the size is not determined. A PU may contain multiple TUs, one TU may also span multiple PUs.
H.266 In this basis, in addition to the four-fork tree, it adds three trigemia and binary tree division.
Thinking: Can I divide irregular shape? For example, triangles, circles, ellipses, hexagons, etc.
4. Inframe prediction:
The 4 × 4 and 8 × 8 blocks in H.264 include nine prediction modes, 16 × 16 blocks contain four prediction modes;
VP9 has 10 intra prediction modes;
H.265 has 33 intra-frame angle prediction modes + DC (upper, left average) + PLANER; compared with H.264 / AVC, H.265 / HEVC increases the boundary pixel of the lower left block as a reference for the current block. ;
H.266 has 65 intra-frame brightness angle prediction modes, actually 65 + 10 + 10 = 85 species, according to the aspect ratio; increase ISP (further division technology to block); PDPC technology, combined with unfiltered wave Reference pixel and filtered reference pixels, increase MIP mode; CCLM mode;
Note: The Planar mode is suitable for the area where the pixel value is slow, it uses two linear filters in the horizontal and vertical directions, and the average of the two is used as the predicted value of the current block pixel. The DC mode is suitable for large-area flat regions, and the current block predictive value can be obtained from the average of the reference pixels on the left and above. The angle mode is mainly used for textures in different directions in video content.
5. Inter-frame prediction:
Frame type structure: H.265 uses Hieracle-B structure
MV Accuracy: H.265 is pixel precision (chroma) and uses more adjacent pixel points for sub-pixel precision interpolation. Prediction mode: SKIP, Direct, Merge (5 candidate MV), AMVP (2 candidate MV).
H.266 increases the pixel precision;
The VP9 inter-frame prediction uses the pixel to perform motion compensation, and there is a non-display frame as a reference frame, and there is an average bidirectional prediction on the frame.
Reference list:
H.265 With two reference lists, each of them has 16 reference items, but the maximum number of unique pictures is 8.
The Merger mode in H.266 has 6 Candidate, which adds TMVP, HMVP changes relative to H.265.
6. Transformation
H.264 integer DCT 4x4 8x8; Hadama transform
Both VP9 and HEVC support the size of the 4x4-32x32. DCT in the intra encoded macroblock, one or both of the vertical and horizontal transform paths will be DST
HEVC 4X4 DST; Transform_skip Mode: Transform_skip_flag, this mode has a good effect on text desktop video; RQT technology based on adaptive transform technology of four-tree; no Hadama transform
HEVC internal bit depth increases: in order to ensure intermediate prediction, transformation, and internal bits during the quantization process to achieve better compression performance
HEVC only adopted a transformation of 4 o'clock DST7 for intra predictive residuals, and DCT2 was still adopted for other sizes and inter predictive residuals;
H.266 has a non-separable secondary transform LFNST; the predicted residual transformation can better adapt to the statistical characteristics of the predicted residual dynamic change, and significantly further increase the transform gain. Sub-block transform, SBT) for interframe conversion technology
7. Entropy encoding:
H.264 is compressed by integer discrete cosine transform (DCT), CABAC compression (non-destructive, CABAC is also short-frequency data short code, to low frequency data long code. At the same time, it is compressed according to the context-dependent correlation), between two I frames An image sequence GOP.
VP9 supports four transformations: 32x32, 16x16, 8x8 and 4x4. These transformations are the approximate integers of DCT as many other codes. In the intra encoded macroblock, one or both of the vertical and horizontal transform paths will be DST (discrete sinusoidal transformation).
HEVC's entropy coding uses two arithmetic coding: Cabac and Cavlc. CAVLC is mainly used for encoding SEI, parameter set, title, etc., all of which are all coded in all data and syntax elements.
H.265: Zigzag scan: ACS technology, vertical scan, horizontal scan, diagonal scan.
8. Filter:
H.265 added SAO
H.266 added ALF, brightness 7x7, chromaticity 5x5
VP9 can choose from three different sub-pixel interpolation filters per block:
Normal eight-pixel / smoothing eight-pixel, can perform smoothing or fuzzy prediction / sharp 8 pixels, can perform sharp prediction
9. Acceleration technology
H.265 adds parallel tool sets such as Tile and WPP to increase coding speed
TILE divides the image into a TILE block for a rectangular area as a basic parallel unit, which can simultaneously include multiple Tiles and some TILEs containing multiple sloph of the plurality of SLICEs.
WPP: Full called Wavefront Parallel Process, basic encoding units based on LCU behavior.
Taking a line of LCU blocks as a basic parallel unit, each line of LCU is a sub-code stream
10. Others
VP9 optimizes the 8th pixel accuracy of the motion vector, three switchable sub-pixel interpolation filters, reference motion vectors, entropy coding, loop filter, ADST, DCT, etc.
H.264 Level: Description of the video, the higher the Level, the higher the code rate, resolution, and fps
H.266: Chromatic Union Code JCCR
HEVC IBDI Technology
Our other product:
