[Image-SIG] Need 12 or 16 bits/pixel images

[David Ascher]

On Sun, 30 May 1999, Rob Hooft wrote:

> I have a similar application. Scaling the image to 8 bits is really
> the only sensible thing to do, as the human eye cannot differentiate
> much more anyway, and there are AFAIK no regular graphics cards that
> can do more than 8 bit greyscale.

FWIW, I can think of applications where > 8 bits matters, both in cases
where perception is involved and cases where the images contain data which
is to be analyzed by non-human processes (as I suspect is the case in
Roger's app).

If you do find you have to settle for 8-bits, depending on the case, you
might find that some other compressive transform besides scaling is more
appropriate (e.g. by knowing something about the noise in the imaging
device, information loss might be reduced by compressing e.g. the low
values more than the high values).  The image won't be "veridical", but
then again, neither are the CCD, the monitor and the eye.

I have no answer to Roger's original question, alas.  

--david "vision scientist" ascher