[转载] ISO 14496-10 (Video) - Advanced Video Coding (AVC) (from doom9 )

With AVC/H.264 the MPEG-4 Standard defines one of the newest and technically best available, state-of-the-art Video Coding Formats

The AVC/H.264 Video Coding Standard was together finalized and identically specified in 2003 by two Groups, the MPEG (Moving Pictures Experts Group) from ISO and the VCEG (Video Coding Experts Group) from ITU (International Telecommunication Union), a suborganisation of the United Nations (UNO), which also standardised the H.263 format (mainly used in video conference software now)
The AVC/H.264 Standard itself was developed by the Joint Video Team (JVT), which included experts from both MPEG and VCEG

Looking from the MPEG side the standard is called MPEG-4 Part 10 (ISO 14496-10), looking from the ITU side, it is called H.264 (the ITU document number), by which the format is widely known already
As "official" title for the new standard Advanced Video Coding (AVC) was chosen by MPEG - as video counterpart to the Advanced Audio Coding (AAC) audio format


AVC/H.264 Profiles

The AVC/H.264 standard defines four different Profiles: Baseline, Main, Extended and High Profile (which themselves are subdivided into Levels):

- Baseline Profile offers I/P-Frames, supports progressive and CAVLC only
- Extended Profile offers I/P/B/SP/SI-Frames, supports progressive and CAVLC only
- Main Profile offers I/P/B-Frames, supports progressive and interlaced, and offers CAVLC or CABAC
- High Profile (aka FRExt) adds to Main Profile: 8x8 intra prediction, custom quants, lossless video coding, more yuv formats (4:4:4...)

It seems the most usable profile for DVD Backups is the High Profile with maybe the following tools (also check out the tool description of MPEG-4 ASP as all, except GMC, are available in AVC too):

CAVLC/CABAC:
AVC/H.264 defines two, more advanced tools for entropy coding of the bitstream syntax (macroblock-type, motionvectors + reference-index...) than MPEG-4 ASP: Context-Adaptive Variable Length Coding (CAVLC) and Context-Adaptive Binary Arithmetic Coding (CABAC)
CABAC, compared to CAVLC (aka UVLC) which is the default method in AVC/H.264, is a more powerful compression method, being said to bring down the bitrate additonally by about 10-15% (especially on high bitrates). CABAC (as CAVLC) is a lossless method and therefore will never hurt the quality, but will slow down encoding and decoding.

Loop/Deblocking Filter:
in contrary to prefiltering (for example via avisynth, done on the input), or postprocessing/filtering (via the decoder, done on the final output), LoopFiltering is applied during the encoding process on every single frame, after it got encoded, but before it gets used as reference for the following frames. This helps avoiding blocking artifacts, especially on low bitrates, but will slow down en/decoding

Variable Block Sizes/Macroblock Partitions:
in contrary to MPEG-4 ASP (where, only with Inter4V/4MV, the Block Sizes can varry between 16x16 and 8x8 pixels), AVC/H.264 offers for Motion Search Precision the division of a macroblock down to 4x4 pixels (including steps like 8x4...). The Block Size is adaptive/variable, a good encoder will be smart enough to decide which one is best to use in every specific macroblock

Multiple Reference Frames:
in contrary to MPEG-4 ASP (which only allows using the frame before the actual frame as reference), AVC/H.264 offers choosing from multiple ones for inter motion search, which means the codec can decide whether he wants to simply refer to the previous frame (like in ASP) or even to a frame before that. Because of that (eg a P-Frame can refer to a frame before the latest I-Frame) a new frametype had to be introduced: IDR-Frames, which are I-Frames before which no following frame is allowed to refer to. Allowing multiple reference frames will slow down encoding and decoding and cutting will be only possible at IDR-Frames

Weighted Prediction:
With Weigthed Prediction there can be weights applied to a reference frame (eg you can scale (brightness-wise) a previous picture). This helps especially whenever there are fades, where the subsequent picture is very similar to the previous one except that it is darker. WP will not help with cross-fades (eg a fade from one scene to another)

Rate Distortion Optimisation (RDO):
RDO allows the encoder to make the most efficient coding decisions whenever it has to choose between different choices (for example when it comes to inter/intra decisions, motion search...)
RDO is not a tool defined by the AVC/H.264 specs, but it's a new decision making approach which was first introduced by the H.264 reference software. Other codecs can also make use of RDO, like XviD's VHQ Mode enables RDO already for example


An overview of AVC/H.264 compared to other popular video coding formats:


available AVC/H.264 Codecs

AVC/H.264 implementations are available atm already from x264, Nero, Apple, Sorenson, Elecard, Moonlight, VSS, mpegable, Envivio, Hdot264 ( binary), DSPR, JM (reference software) ( binary), ffmpeg, Philips, FastVDO, Skal, Sony and many more

Encoders

- x264: the first publically available High Profile encoder, opensource (GPL) ( Source), available for VFW: x264vfw, ffdshow (output .avi), as commandline: x264cli (outputs .mp4, raw), mencoder (outputs raw, .avi) (Doom9's MeGUI) or ffmpeg
x264 supports 2pass, CABAC, Loop, multiple B-Frames, B-References, multiple Reference Frames, 4x4 P-Frame, 8x8 B-Frame Blocksizes, anamorphic signalling and High Profile: 8x8 dct and intra prediction, lossless and custom quant matrices
- NeroDigital AVC: useable in Nero Recode2, outputs .mp4
ND AVC supports 2pass, CABAC, (adaptive) Loop, multiple B-Frames, mulitple Reference Frames, weighted prediction, 8x8 P-Frame Blocksizes, 16x16 B-Frame Blocksizes, Adaptive Quant. (Psy High)
- Sorenson: useable in Sorenson Squeeze 4, outputs .mp4,
Sorenson supports 2pass, max 2 B-Frames, B-References, Loop and multiple Slices
- Apple: useable in Quicktime 7, outputs .mp4, .3gp and .mov, totally slow
uses 2pass, max 1 B-frame, Loop (0,0), P8x8,B8x8,I4x4, Adapt. Quant, 5 Slices, no CABAC, no Weighted Pred., no multi Ref.
- JM: The AVC Reference Software offers in v9.3 Main and High Profile: B/SP-Frames, CABAC, Loop Filter, 4x4 Blocksizes, multiple Reference Frames, Adaptive Quant, Error Resilience, RDO, Lossless Coding, Custom Quants, Rate Control aso...
- Hdot264: opensource (GPL) VFW version of the reference software by doom9 member charact3r, still based on a very old version of the reference (JM 4.0c)
- VSS: free preview VFW Encoder (limited to 5 days), based on the reference encoder
- Elecard: useable in Elecard Mobile Converter, outputs .mp4 and MainConcept's v2 encoder, outputs .264 and .mpg PS/TS
not publically available anymore:
- Moonlight: useable in Moonlight's OneClick Compressor v1.1 and CyberLink's PowerEncoder, outputs .mpg
Moonlight supports 1pass (VBR/CBR/Fixed Quants), CABAC, Loop, 2 B-Frames, 8x8 P-Frame Sizes, Adapt. Quant, PAR, Interlacing
- MainConcept: was useable in the v1 encoder (adds a watermark), outputs .264 and .mpg PS/TS
1pass (CBR/VBR/fixed Quants), P-Frame Reorder, CABAC, Loop, Multiple B-Vops, Multiple Ref, 4x4 P-Frame Sizes, PAR, RDO
- mpegable: offered for some time a free VFW Encoder (not based on the reference), doesnt handle YV12
mpegable supports 1pass (fixed quants) uses P-Frames only, 8x8 P-Frame Blocksizes, CAVLC only, Loop
- Envivio: useable in 4Coder, outputs .mp4

Decoders ( comparison)

- ffmpeg: opensource (LGPL), used e.g. in ffdshow (VFW and DShow decoder), mplayer and VideoLAN
supports B-Frames, B-References, CABAC, Loop, Weighted Prediction and High Profile (8x8 dct and intra prediction, lossless)
- Apple: AVC decoding inside Quicktime 7, supports .mp4/.mov, very slow
supports only 1 B-Frame, CABAC, Loop but no mixed references, multiple B-frames and no interlacing
- NeroDigital AVC: DShow Decoder and .mp4 Parser coming with Recode2
supports Main and High Profile
- VSS: preview VFW Decoder (limited to 5 days) and a DShow Decoder (limited to 30 days)
VSS DShow supports .avi (with VSSH and H264 fourcc), CABAC, Loop, B-Frames
- Elecard: available in Elecard's MPEG Player v4.0 and MainConcept's v2 encoder
- Envivio: not freely available AVC DShow decoder called EnvivioTV, handling AVC in .mp4 (since 2.0, current version: 2-1-181)
- Philips: DShow AVC decoder freely available in the AVC Alliance player (handles raw AVC only)
- FastVDO: time limited (5 minutes per video) High Profile DShow Decoder
- Pegasus: not really compliant DShow AVC decoder available here
- Basic AVC Decoder in C, for an university project available here
not publically available anymore:
- Moonlight: DShow decoder/Parser handling AVC in .mpg, .mp4 and .264 available in Moonlight's MPEG Player v3.0
supports Main and High Profile
- MainConcept: the v1 preview offered a free DShow AVC decoder (adds watermark) and Parser handling AVC as .mpg PS/TS
- mpegable: offered for some time a free VFW decoder (usable also in DShow), supports .avi (with DAVC fourcc)


Sample content

NeroDigital: mp4, mp4
Sorenson: mp4
AVC Alliance: raw
Moonlight: raw/medium bitrates, raw/low bitrates, raw, mpg
FastVDO: raw/high profile
Apple: mov
Videosoft: avi, avi/new, avi/old
Lead: ogm


current issues with AVC/H.264

- interoperability: most implementations support different container formats atm:
.mp4: which is the container of AVC defined in the MPEG-4 Standard (ISO 14496-15) and supported by Apple, Nero, Sorenson, Envivio, Elecard/Moonlight and x264 atm
.mpg PS/TS: which are the containers of AVC defined in the MPEG-2 Standard (ISO 13818-1, AMD3) and supported by Mainconcept and Elecard/Moonlight atm
.avi: using AVC-in-AVI is nowhere standardized and therefore already causes incompatibilies. The limitations of AVI and VFW (eg regarding b-frames or arbitrary frame coding orders), together with the necessary hacks caused by these two formats, hinder the full implementation of all possible features AVC offers and therefore harm the possible quality or at least the speed of the development, the interoperability and therefore also the competition. AVI is currently used by VSS and x264 (mencoder and vfw)
.264/.h264: the raw bitstream not stored in a container. output for example by the reference, x264cli, mencoder and mainconcept

- speed: some current implementations are pretty slow
still x264 and NeroDigital's AVC encoder seems to offer already a nice speed and quality. But this doesnt change the fact that AVC is a very advanced video coding format and therefore encoding and decoding on old CPU's can be very time consuming


MPEG-4 AVC/H.264 on Hardware - HD-DVD/Blu-ray

the DVD Forum and the Blu-ray Disc Association are currently working on successors for the DVD format, supporting High Definition content (simply larger picture sizes than current DVD): HD-DVD and BD-ROM

As reported here MPEG-4 AVC/H.264 will be mandatory for HD-DVD
Blu-ray has also included MPEG-4 AVC/H.264 as written here

It is therefore very likely that AVC/H.264 will be THE upcoming video format, which will be widely used and supported, like it is the case with MPEG-2 (used in DVD) today


further documentation

Read more about the MPEG-4 AVC/H.264 here for a detailed overview, summarized info here or here
a list of available implementations
The AVC Verification Test Results can be found here
The whole specs of AVC/H.264 can be downloaded here (Draft from the 7-14 March 2003)
Technical Info about Blu-ray is available here