Re: Subtracting G channel from R and B may increase lossless compression?


"Frank Mertens" <frank@xxxxxxxxxxx> wrote in message
news:f3bpsm$ahn$1@xxxxxxxxxxxxxxxxxxxxxxx
LiloLilo wrote:
Hi all,

anybody knows how subtracting G channel from R and from B could
increase lossless compression? I did some tests on some images and found
out that this is true but not for all images. Images with more noise
compress worst.
Do you think that subtracting for example the R channel or the B channel
instead of G would be different?

Best Regards, Danilo

I tried some color differencing of type G, G-R, G-B, but found it
wouldn't reduce the entropy of the separate channels very much. The
basic problem was that the difference of two neighbor pixels in the same
channel has shown to be much lower than the channel differences of the
same pixel. Which actually makes sense, because the different spectral
parts should show something quite different, if not the whole picture is
much to grayish.
For me, LOCO prediction of the separate channels without color transform
and feeding it into a entropy coder achieved highest ratios. Beware that
the difference G-R costs a bit more than just R, because lossless
addition/subtraction of 2 n-bit words requires a (n+1)-bit word as result.


usually, it is R-G and not G-R.
also note, that this is usually combined with predicting pixels from
adjacent ones (since the color differences typically follow similar
gradiants to the original values).

also note that greyish-ness is a major factor of many images as well.

after all, life is mostly just shades of grey, and these colors, they are
largely tacked on and arbitrary (arbitrary names, arbitrary rules, one thing
one color, another another, so one remembers what thing is what color, but
in reality all this is largely arbitrary...).


Regards, Frank.


.

Relevant Pages

  • Re: Subtracting G channel from R and B may increase lossless compression?
    ... increase lossless compression? ... out that this is true but not for all images. ... Do you think that subtracting for example the R channel or the B channel ...
    (comp.compression)
  • Re: Color Image Formation
    ... I think I have a better understanding of the importance of the reflection of the light. ... In the particular application I'm interested in (satellite images), I'm still not convinced this occurs a lot. ... (u1 corresponds to the red channel, and u2 corresponds to the green ... Snow is white. ...
    (sci.image.processing)
  • Re: Subtracting G channel from R and B may increase lossless compression?
    ... increase lossless compression? ... out that this is true but not for all images. ... Do you think that subtracting for example the R channel or the B channel ... other lossless color transforms may be possible as well (as many argue ...
    (comp.compression)
  • Re: Choice between channels
    ... from simple RGB images ... High frequency contrast is noise. ... Say I filter each channel using a 10-pixel wide spatial filter. ...
    (comp.graphics.algorithms)
  • Re: Images Turn Blue
    ... Yeah, there is a layer of bugs in play here, like the skins of an onion. ... This is a "channel inversion" problem, and we have had that for years in Mac ... But if the filters were able to fin any vector images, ... Thus they contain bitmap and vector art. ...
    (microsoft.public.mac.office.word)