Complexity is not the only reason to select 8x8
as the block size. DCT mimics the behavior of KLT for AR-1 sequences (with rho
near to 1) using sinusoidal bases. For DCT to work well in compression, the data
should better be as close to an AR-1 (rho near to 1) as possible. Although it
depends on the resolution (and other things) of the image that is considered, a
8x8 block for a VGA sized image is usually suitable for hat purpose. For sizes
larger than 8x8, the AR-1 property gets spoiled. However, obviously, this is not
true for images with several megapixels of resolution...
Omer
----- Original Message -----
From: venkatesh viswanathan
To: Lakshman ; n...@rediffmail.com ; getsunil_6 ;
i...@yahoogroups.com
Sent: Tuesday, August 24, 2004 6:36 PM
Subject: RE: [imagedsp] Size of the macro block in DCT
It is a tradeoff between quality and complexity as mentioned below.Ideally
it would be nice to have smaller blocks but that also means a lot of motion
vectors to be coded (involves searching and coding).Hence 16 x 16 would have
been an optimal solution considering the complexity. Lakshman
<l...@yahoo.com> wrote: Hi, all
As we know, the computational complexity of DCT increases with
increase in block size. If you plot MSE Vs Block Size, there is no
significant gain in choosing block sizes larger than 8x8.
Cheers -Lakshman --- Nalin Pithwa <n...@rediffmail.com>
wrote:
> Hi, > > It is most probably due to the
following reason: We > know that motion > compensation is used
in MPEG 4 and H.263/H.264. For > instance, the energy in > the
residual is reduced by motion compensating each > 16 x 16 macroblock. > Motion compensating each 8 x 8 (instead of each 16 x > 16
macroblock) reduces > the residual energy further. These examples
show > that smaller motion > compensation block sizes can
produce better motion > compensation results. > However, a
smaller block size leads to increased > complexity (more search > operations must be carried out) and an increase in > the number
of motion > vectors that need to be transmitted. Sending each >
motion vector requires > bits to be sent and the extra overhead for
vectors > may outweigh the benefit > of reduced residual
energy. > > Cheers, > Nalin Pithwa > >
-----Original Message----- > From: getsunil_6
[mailto:g...@yahoo.co.in] > Sent: Tuesday, August 17, 2004 4:59 PM > To: i...@yahoogroups.com > Subject: [imagedsp] Size of the macro
block in DCT> Hello all, > > I am having a small doubt
regarding DCT calculation. > Why we are going > for 8 x 8
macroblocks in DCT but not higher or lower > than that? > Please
let me know this. > > Thanks in advance > Sunil.>
_____________________________________ > Note: If you do a simple "reply"
with your email > client, only the author of > this message will
receive your answer. You need to > do a "reply all" if you > want your answer to be distributed to the entire > group. > > _____________________________________ > About this
discussion group: > > To Join: i...@yahoogroups.com > > To Post: i...@yahoogroups.com > > To
Leave: i...@yahoogroups.com > > Archives:
http://www.yahoogroups.com/group/imagedsp > > More Groups:
http://www.dsprelated.com > Yahoo! Groups
Links ____________________________________________________________________ Note: If you do a simple "reply" with your email client, only the author of
this message will receive your answer. You need to do a "reply all" if you
want your answer to be distributed to the entire group.
_____________________________________ About this discussion group:
To Join: i...@yahoogroups.com
To Post:
i...@yahoogroups.com
To Leave: i...@yahoogroups.com
Archives: http://www.yahoogroups.com/group/imagedsp
More Groups:
http://www.dsprelated.com
|