content:
1. Code frame difference
2. Comparison of compression performance
3. Each module technology difference summary
4. Block division structure
5. Inframe prediction
6. Inter-frame prediction
7. Device filtering
8.SAO filter
9.TILE
10.Wpp
11.dependent Slice
12. Other technology
Be
1. The main difference between H.264 and H.265 H.265 is still mixed, codec domain H.264 is basically,
The main difference is:
Ø Code block division structure: The recursive structure of Cu (CodingUnit), PRIXIT and TU (TransformUnit).
Ø Basic details: There are many differences in the internal details of each function block.
Ø Parallel Tools: Added parallel tool sets such as Tile and WPP to improve coding speed
Ø Filter: Added SAO (Sample Adaptive Offset filtering module after being filtered)
Fig.h.265 frame map
Be
2. Compression performance Compare PSNR calculation mode:
H.265 / HEVC HM-9.0 and H.264 JM-18.4 BD-RATE Comparison:
Allintra Case: 22%
Randomaccess Case: 34%
Lowdelay Case: 37%
Be
3. Each module technology difference summary
4. Block division structure In H.265, the size of the macroblock is extended from H.264 to 64 × 64 to facilitate the compression of high resolution video.
At the same time, a more flexible encoding structure is used to improve coding efficiency.
Includes coding units, predictUnit, and transformunit.
As shown below:
in:
The encoding unit is similar to the concept of macroblock in H.264 / AVC, for encoding the process.
The prediction unit is a basic unit for prediction,
The transform unit is a basic unit for conversion and quantization.
The separation of these three units makes the transformation, prediction, and encoding various processing links,
It is also conducive to the division of each link to more in line with the texture characteristics of the video image.
It is conducive to the completion of their respective functions more optimized.
RQT is an adaptive transform technology that extends and expands the extension and expansion of ABT (AdaptiveBlock-size transform) technology in H.264 / AVC.
For inter-frame encoding, it allows the size of the transform block to adaptive adjustments according to the size of the motion compensation block;
For intra coding, it allows the size of the transform block to adaptive adjustments according to the characteristics of the intra predictive residual.
Big block transformations can provide better energy concentrations with respect to small blocks, and can save more image details after quantification, but on the other hand, there will be more ringing after quantification. effect.
Therefore, according to the characteristics of the current block signal, the adaptive selection variable block size, as shown in the figure below, can be obtained by the excellent energy concentration, the degree retention level, and the rings effect of the image.
Flexible block structure diagram
Be
5. The intra prediction mode is in nature H.265 is based on the prediction direction of H.264, adding more forecasting directions.
H.265: All size Cu blocks, 35 prediction directions in brightness, 5 prediction directions
H.264: 9 directions of brightness 4x4 block, 8x8 blocks in 9 directions, 16x16 blocks 4 orientation, 4 directions
The intra prediction direction of H.264:
The intra prediction direction of H.265:
Be
6. Inter Bloom prediction Essentially H.265 is a number of tap coefficients, changing the tap coefficient value, and increasing the number of exercise vector predictive values, to achieve the purpose of reducing predictive residuals.
H.265 is the same as H.264 is brightness to 1/4, chromaticity to 1/8 precision, but the interpolation filter tip length and coefficient are different.
H.265 increases the number of motion vector predicted values, while H.264 predicted value only one
H.265 airspace candidate:
H.265 Time Domain Common Position Candle
Be
7. The deblocking filter Essentially H.265's deck filtering process is consistent with the H.264's deck filtering process, which has made the following most significant changes:
Ø Filter boundary: H.264 minimum to 4x4 boundary filtering; while H.265 adapts to the filter edge of the latest CU, PU and TU division structure, minimum filtering boundary is 8x8,
Ø Filter order: The vertical boundary is used in the H264 prior macro block, and the horizontal boundary of the current macroblock; and the vertical boundary of the H.265 first frame, the horizontal boundary of the whole frame
The ALF is within the codecline, after DEBLOCK and SAO,
Used to restore the reconstructed image to achieve a wifter difference (MSE) between the reconstruction image and the original image.
The factor of the ALF is in frame-level computing and transmission, and ALF can be applied to the entire frame.
ALF can also be performed for partial regions based on block or quadtree.
If it is an ALF based on the partial area, additional information indicating the area information must also be passed.
8. Sample AdaptiveOffset filtering SAO (Sample Adaptive Offset) Filtering is actually classified according to different templates, and compensates for each classification pixel after the deblocking filtering. For BO (BAND Offset) and EO (Edge Offset).
BO Category:
EO classification module:
SAO is in the codeca, after DEBLOCK, by the classification of the reconstructed image, add an offset to each type of image pixel value, reducing the purpose of reducing distortion, thereby increasing the compression ratio, reducing the code stream.
After SAO, the average can reduce the code stream of 2% to 6%, while the performance consumption of the encoder and decoder increases by about 2%.
9. TILE TILE: Split the image into a rectangular area.
Its main purpose is to enhance parallel processing performance.
Each Tile region is equivalent to a sub-image, which can be coded in units of LCU blocks.
A Tile block is a basic parallel unit, each Tile is a subcode stream.
10. WPP
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
11. Dependentslice
Dependent Slice: This technology can be understood as a data division of the original Slice NALU to make it suitable for more flexible packaging methods.
The schematic of Slice and Dependent Slice is as follows
12. Other related technologies
Ø Transform_skip mode: transfer_skip_flag, this mode does not transform, but it is quantified, this mode has a better effect on text desktop video
Ø Internal bit depth increase: in order to ensure intermediate prediction, transformation, and internal bits in the quantization process to achieve better compression performance: https://blog.csdn.net/firell/Article/details/77827156
Our other product:
