Re: [fluid-dev] Purpose of dither?

[Mihail Zenkov]

On Fri, 11 May 2007 12:54:45 -0700 (PDT)
Z F <address@hidden> wrote:

> 
> --- Miguel Lobo <address@hidden> wrote:
> 
> > >
> > > It does not matter if you add random float to a float and truncate
> > or
> > > trancate and add a random bit, the outcome will be exactly the same
> > > sequence of bits so it does not matter, you can always add dither
> > > later.
> > >
> > > Write it out...
> > >
> > > trun(signal + noise) = trun(signal) + noise'
> > >
> > > notice that noise' is not the same as noise, but the equality holds
> > > so one can do what can not be done -- remove undesired harmonic
> > > after truncation.
> > 
> > 
> > The problem is that there is no way of calculating noise' after the
> > truncation.  You would need the information that was lost in the
> > truncation.
> > 
> > I think a visual example might be easier to understand.  In the
> > Wikipedia
> > page for dither http://en.wikipedia.org/wiki/Dither they have a
> > series of
> > pictures of a cat.  The first one is the original high-depth
> > photograph.  In
> > the second picture, truncation has been applied without dithering and
> > you
> > can see big flat areas of uniform color.  How can you know which was
> > the
> > original shade of a pixel in those flat areas?  You can't.  In the
> > third
> > picture, truncation has been applied with dithering.  You can see
> > that
> > although the palette of the third picture is the same as that of the
> > second,
> > the "average" color of each area has been much better preserved. 
> > There are
> > more bright pixels where the original picture was brighter, and fewer
> > where
> > it was darker.  To do that you need the information in the original
> > picture.
> 
> 
> No, you do not... :) Ever heard of texture synthesis? All you need
> is statistical properties, not the signal itself to generate which
> looks like a cat. So, it can be done without the original :) Since the
> main features of the signal are preserved after truncation (otherwise
> it would have been useless), it should be enough to synthesize noise'.

Yes, we can truncate original picture to three chars  - 'cat', but it
have much less useful info about original object ;)
Any truncate incorrigible, with dithering we have much less loss, then
without. Any synthesis - fake, it improve quality, but lack of
similarity with original.