On Tue, Apr 13, 2010 at 2:16 AM, Guy K. Kloss <g.kl...@massey.ac.nz> wrote: > On Mon, 12 Apr 2010 23:32:32 Sebastian Haase wrote: >> are you sure it makes even sense to save a 16-bit RGB image ? This is >> not meant as an excuse for PIL to not support it, >> but 16mio colors should likely be enough for any application (i.e. >> 8bit per R,G and B) > > It does make sense. Absolutely! Maybe not if you are *just* thinking in terms > of final output for an end user, but during the whole > capturing/processing/manipulation phase one can reduce many artifacts > introduced through rounding, etc. Also when images are touched up by changing > lightness or contrast images tend to expose "banding" quite severely. > > This and other effects are also the background behind the increasing > popularity of HDR (High Dynamic Range) imaging. Particularly in scientific > imaging subtle differences are much preserved this way. And it looked like > Dan's image is the result of a microscopic picture, or something like that, > with low contrast. So it could strongly benefit from a change in lightness and > contrast. > > Higher channel bit depth are important for these cases, and even more to keep > up with needed capabilities for the future! > > Guy
I was talking only about the information content captured by physically collecting photons from a film with very short exposure times. That is, the "capturing" phase; for what you call the "processing/manipulation" phase I would always convert to single-precision float (numpy.float) if that fits into memory. This way you are save if values get negative or intermittently very small, for example. I'm just a bit confused here, because I am used to collecting gray scale images, not RGB images, (from a cooled CCD on a microscope). Those I also save as unsigned 16 bit integers. BTW, the original example image looked also pretty "gray" to me, could you tell the scanner to use 16-bit gray, maybe then the (physical) scanning quality might even be better - don't know, just wild guess... Sebastian _______________________________________________ Image-SIG maillist - Image-SIG@python.org http://mail.python.org/mailman/listinfo/image-sig