Re: [APML]: Stacking images in photoshop

[Wil Milan <wmilan@airdigital.com>]

Jerry Lodriguss wrote:

> A question about sofware such as MIRA. Is there a limit on the file size
> you can work on?  My understanding was that with a lot of these CCD imaging
> programs, large files sizes (like 24 mb -72mb) are not supported.

Mira isn't really a CCD imaging program but a general-purpose scientific-imaging
package. A lot of the application examples deal with microscopy, X-ray imaging,
and other scientific applications other than astronomy or CCD imaging
specifically.

Mira can manipulate files up to 2 gigabytes in size, which should cover just
about any conceivable imaging requirement. It uses its own virtual-memory paging
scheme so that if you have enough free disk space it can simulate up to 3.5
gigabytes of RAM. Of course, having lots of real RAM speeds things up a lot.

> Stacking 8 bit per channel originals still yeilds 8 bit composites, don't
> they?  I mean, you seem to imply that stacking two 8-bit per channel images
> will yeild 16-bits per channel, and I don't think that is correct. Of
> course, like Homer Simpson, sometimes, "Jerry no function beer well
> without," and it is pretty early for me. <G>

Stacking two 8-bit images *may* result in values which can still fit in 8 bits,
but it usually doesn't.

With 8 bits per channel you can represent 256 discrete levels per color channel
per pixel. If you add two pixels from two different images together and each
pixel has a value of, say, 97, the resultant value will be 194, which will still
fit within 8 bits. But if the two pixels being added have values of 157 and 206,
the sum of the two will be 363, which will not fit in 8 bits. The result will be
to clip off the value at 255, which results in a distortion of the color signal.
Bright areas will lose all detail and to completely solid, the digital equivalent
of burned-out white on film.

To avoid such clipping, the common thing to do is to divide all the pixel values
in each image by 1/2 to stack two images, by 1/4 for four images, etc. That
guarantees that the summed values will not overrun the 8-bit range limitation,
but each time the values are divided there is some loss of resultion. Dividing
each value in half in effect compresses 8 bits into 7 bits, so in effect you are
reducing the color resolution on each channel to 7 bits, not 8.

Using an image-processing package which allows 16-bit or larger pixel values
avoids this by capturing the full value of multiple summed images. Some
scientific-imaging packages also allow pixels to be respresented by 16- or 32-bit
floating-point values, which has advantages for some kinds of numerical
processing.

> And this is really pretty much of a moot point anyway for those of us who
> shoot film, because all of the scanners may scan at more than 8 bits per
> channel, but you can't get more than that into a program because of the
> limits of the scanner software, as far as I know.

True, but stacking multiple images typically results in pixel values which
require more than 8 bits per pixel.

> Another question, do these scientific programs even support film scanners?

TIFF format is commonly supported by most imaging packages, and every scanner
driver I've seen can generate TIFF files.

> How would you get more than 8 bit per channel data into them from film?

If nothing else then by stacking, as discussed above. But I'm not sure there
aren't scanner drivers capable of more than 8 bits per pixel output (if there
aren't, here's a golder opportunity for someone). And of course most CCD cameras
generate 12 or 16 bits per channel.

> I've always wanted to take advantage of such features as offered in the
> scientific programs, but the above mentioned questions have always held me
> back.

Sophisticated image processing certainly does wonders with CCD images. I'm just
getting started with applying such processing to film images, but the
possibilities are certainly much broader than the limited processing one can do
with PhotoShop, PhotoPaint and ilk.

One consideration with applying such processing to large film images: It does eat
up RAM and CPU capacity. My 150-MHz Pentium and 80 Mb RAM is barely adequate, and
I've been thinking what I really need is a 300-MHz Pentium II and twice as much
RAM. A dual-Pentium machine would be even better, and four Pentiums and 256 meg
of RAM would not be too much. One of the servers we use at my office is a
6-Pentium machine with 512 meg of RAM, and at times I've thought about what a
zippy image-tweaking machine that would make. Then there are those nifty Sun
workstations...  :-)

In truth you don't really *have* to have the latest killer Pentium to run this
stuff, but the more juice the better. And I definitely wouldn't try it on a 486,
not unless you have hours to spare.


Wil M., waiting for the 64-bit Intel Merced to come out :-)
--
"When I consider your heavens, the work of your fingers, the moon and the stars
which You have set in place, what is Man that you are mindful of him, or the son
of Man that you care for him?"  -- Psalm 8