[XviD-devel] why DCT image/video compression?

[skal]

	Hi all,

On Wed, 2004-09-22 at 09:01, Christoph Lampert wrote:
> On Wed, 15 Sep 2004, prabhakar p wrote:
> 
> > Hi Guys,
> 
> >      Hope ur doing fine there.Can anybody let me know, what all the
> >      reasons DCT is preferred over other Transformations in image/video
> >      compression techniques.
> 

	more precisely: DCT is a (good) approximation of
	the optimal Karhunen-Loeve transform that is best
	at compacting and de-correlating signals modelled
	as 1rst-order Markov processes. You wouldn't use
	it for handling radio-activity signals, but for
	video, that's ok.

	bye!
Skal

> Like _what_ other transformations? 
> 
> DCT is a classical approximation of PCA, and very succesful for it 
> computational efficiency and very good energy compression property. 
> DCT has been used for several decades, and people and device know how to 
> work with it. The only alternatives, which only came up recently are 
> 
> * integer approximations of DCT like in H.264 or WMV9 (IIRC). Those are 
>   somehwat less effective, but can be done exact to the bit on PCs, which 
>   avoids error drift. So, it's not really a different approach, only an 
>   implementation better suited to PCs. 
> 
> * wavelet transforms, which might be "the future", but which are 
>   mathematically more difficult, and Motion Estimation has many more 
>   problem than in blockbased DCT. Rumors are that RealVideo uses Wavelets, 
>   but I never saw anything concrete about it. I haven't even heard rumors 
>   about hardware for those. 
> 
> DCT was first and works well, and so far, it's the method of choice (i.e.  
> MPEG2, MPEG4, JPEG, DV), until the new standards really kick in (if they
> ever do).
-- 
skal <skal at planet-d.net>