| Author |
Message |
roger
Joined: 27 Aug 2003
Posts: 182
Helped: 7
|
 19 Sep 2003 6:47 why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
I know the concept of object , however , besides this, anything else be
token into account so that mpeg4 get better compression ratio?
does any doc or pdf or website talk about the topic?
I got confused.
teach me! 
|
|
| Back to top |
|
 |
ramesh
Joined: 19 Jan 2003
Posts: 1260
Helped: 5
|
19 Sep 2003 8:38 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
These standards incorporate Digital Video Compression and coding techniques. Depending on the coding techniques used the compression ratio and the quality of the reproduced video signal varies.
Mpeg 1 & 2 incorporate DCT (Discrete Cosine Transform) technique.
Mpeg 4 incorporates Wavelet transorms.
Wavelet is superior compared to DCT in terms of compression ratio and bit-rate.
|
|
| Back to top |
|
|
roger
Joined: 27 Aug 2003
Posts: 182
Helped: 7
|
19 Sep 2003 9:35 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
I see wavelet used in mpeg4 only in VTC(visual texture coding)
wavelet is superior to dct is well known in still image coding,however
if mpeg4 use wavelet rather than dct the problem occurred
" cannot back compatiable to its old version mpeg1,2 "
isn't it true?
if I compared the two system, I would like to use same tools
anything else make mpeg4 get better compression ration than mpeg1,2?
teach me
B.R & TKS
|
|
| Back to top |
|
|
roger
Joined: 27 Aug 2003
Posts: 182
Helped: 7
|
23 Sep 2003 7:43 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
| anybody can help me?
|
|
| Back to top |
|
|
eternal_nan
Joined: 10 Mar 2003
Posts: 158
Helped: 15
Location: eternity
|
27 Dec 2003 8:39 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
You are correct wavelet based coding is only used for textures (still
images) in mpeg 4. As I understand, the performance improvement of
mpeg4 comes mainly from two improvements over mpeg2:
1. Object based coding
MPEG4 treats the video scene as a collection of objects and codes
each object type accordingly (wavelts for textures, something else
for still 3d objects and mpeg2 like coding for video)
2. The accuracy to which macro-block motion vectors can be specified
was improved over what mpeg2 allowed.
|
|
| Back to top |
|
|
roger
Joined: 27 Aug 2003
Posts: 182
Helped: 7
|
08 Jun 2004 9:05 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
most of those declare with mpeg-4 solution are still stay in AVS
(Advanced Simple profile), many of them even without B-frame
, with the concept of object, you have to add many headers,
shape coding, even the dct become SA-dct(Shape-adaptive dct), isn't it?
how can that imporve compression ratio?
|
|
| Back to top |
|
|
pimr
Joined: 19 May 2004
Posts: 81
Helped: 6
Location: India
|
16 Jun 2004 7:58 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
| mpeg1 and mpeg-2 use run length encoding whereas mpeg4 uses arithmetic coding this leads to added computatioal coplexity but better compression than the earlier standards,secondly the quantization coefficients are optimized in mpeg4 which further leads to better comression as compared to other two standards.nowadays companies are targeting h.264 or mpeg4-10 which gives even better efficiency than mpeg4-2 or visual profile.
|
|
| Back to top |
|
|
omiga
Joined: 30 Dec 2003
Posts: 36
|
17 Jun 2004 6:59 why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
mpeg-2 is 8x8 block compress
mpeg-4 is 8x8,8x4,4x4 etc block compress,
and inter , inner frame prediction
|
|
| Back to top |
|
|
Google
AdSense
|
17 Jun 2004 6:59 Ads |
|
|
|
|
|
|
| Back to top |
|
|
yousuf
Joined: 11 Aug 2004
Posts: 8
|
11 Aug 2004 13:22 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
Hi,
As a person who has studied the complete MPEG 2 specs (video only) and is going through the MPEG 4 specs now, I have to say that we can sum it up in a few points:
1. As omiga pointed out, in MPEG 4, motion compensation can take place in blocks of 16x16 or 8x8 or 4x4 etc. etc. The pattern is quite flexible. This has a direct effect on the residue which has to be coded using DCT. If the prediction is good then residual DCT is almost zero which saves a lot of bits.
2. Another option in MPEG 4 is Overlapped Block Motion Compensation. I don't have the exact details of how it works but the idea is to get an accurate motion vector match so as to reduce the contents of the residual DCT data. OBMC also has a de-blocking effect which is good for quality.
3. eternal_nan said that they have improved the motion compensation accuracy. In MPEG1 & 2 only 1/2 pixel was allowed, in MPEG 4 1/4 pixel accurate prediction can also be used. This gains from this feature, however, are content dependant and in certain cases 1/4 pixel accurate motion can increase the frame size! An encoder decision, I guess.
4. Another big thing in motion vectors is the use of Global Motion Compensation. They can significantly bring down the bits needed to encode the motion vectors themselves. This is based on the assumption that the encoder could make an intelligent (one hopes) guess about what's happening to the camera, e.g. zooming or panning or rotating etc. Then using multiple warp points, it could define "global" vector(s) and then just encode the residue of the vectors which would take less bits.
5. As pimr said, MPEG1 & 2 use Huffman (we should not say run-length here) where as MPEG4 can use Arithmetic coding. This gives slightly better compression as Huffman is only "integral" whereas Arithmetic is a "fractional" coder.
6. Yet another thing is shape coding. It can mask certain areas of the image so that better prediction can be made. This is, again, geared towards making better matches to reduce the residue that has to be coded.
In the end I would like to say that I am fairly new to MPEG 4 myself so certain points may not be very accurate. Please forgive and correct me where needed.
Later.
|
|
| Back to top |
|
|
roger
Joined: 27 Aug 2003
Posts: 182
Helped: 7
|
16 Aug 2004 9:45 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
Most of the MPEG4 IP,ASIC(not software tools) only support simple
profile,many of them declared advanced simple profile, but most of them boast. because even no B-vop included. As expers had studied
CIF I required : 150k, P required 50k B required 20k
30fps required 150x2+50x8+20x20=1100(k)
without B: 30fps would required 150x2+50x28=1700(k)
The difference's 1700-1100=600(k),
As yousuf said, points 1,2,3 actually improves the compression ratio. But at the same time more MVs had to be transmitted. Original MPEG1,2 use MBT & CBP.
point 4: GMC's too complicated to implement. Think how can encoder
divide natural video to several VO.
point 5: Arithmetic coding is really better than Hiffman coding,but it's
complicated so it's just an option.
I don't think arbitrary shape coding concept can help compression ratio.
And don't forget, error resilience tools contained in simple profile add
many sync headers, Hec headers, reversible VLC are not like standard
huffman, which should be longer. They definitely are foe to compression ratio.
Will the new efficient tools can compensate 600(k) with the loading of
error resilience headers? I doubt
Last edited by roger on 07 Sep 2004 6:25; edited 2 times in total |
|
| Back to top |
|
|
wassa
Joined: 02 Dec 2003
Posts: 29
|
19 Aug 2004 6:40 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
| i have question about it. How much more effort is required in developing the system for mpeg4 as compared to mpeg1
|
|
| Back to top |
|
|
yousuf
Joined: 11 Aug 2004
Posts: 8
|
19 Aug 2004 7:07 Re: why is mpeg4 superior to mpeg1,2 in compression ratio |
|
|
|
|
Hi,
wassa,
The full implementation of MPEG-4 is not likely in the near future by anyone. It has a very broad scope and I have never seen anyone claim that they have implemented full MPEG-4. The specs are over 500 pages and cover a very broad range of sciences. Consider that MPEG-1 was "derived" from H.261 which had just 17 pages for proper implementation.
The common thing is to support only SP (simple profile) or ASP (advanced simple profile) for MPEG-4. That covers only rectangular picture compression in interlaced and progressive mode. The other stuff in MPEG-4, FBA (face and body animation), Meshes, 3D and still texture compression is not included in SP or ASP.
If you go for SP or ASP of MPEG-4, I'd say roughly it is about 10 times more difficult than MPEG-1.
Hope that helps your estimates.
Later.
|
|
| Back to top |
|
|
