P3D Re: scanning slides

[Jonathan Gross <jonathan.gross@xxxxxxxxxxxx>]

Tom Deering wrote:

> I don't know anything about stereo photography, but as a graphic artist who
> uses a computer every day, I can tell you a little about scanning slides.
> The $1500 scanner mentioned above is overkill, unless you are submitting
> your slide for the cover of a magazine.

"Overkill" is application dependent (depends on what you're doin'), and
there is a big difference between the requirements for high quality
monoscopic and stereoscopic images.  And, if I can bring in a previously
unresolved topic, that difference is exactly why JPEG image compression
is not well suited to stereoscopic images, even though it works great on
2D images.

When looking at monoscopic images, both eyes see the same thing. If
there is some fine detail missing, either spatial or chromatic, it's no
big deal because there is nothing to compare it to but one's own
expectations (and that's a whole other topic of discussion).

With stereographs, each eye sees a different image, and the
psycho-physiological based vision system is inherently sensitive to the
slightest differences from one to the other. 

One application that utilizes the sensitivity of retinal disparity (for
other than stereopsis) is to take two photographs separated in time
instead of space; even the slightest visible difference sticks out
glaringly.  Bill Ewald, a researcher at Eastman Kodak and graduate
instructor at the University of Rochester, tells the story of a
challenge he made to his co-workers.  He bet them that they would not be
able to make even the slightest change to the items on his (very messy)
desktop that he would not be able to detect.  On top of an adjacent file
cabinet that had a clear view of his desk, he used tape to carefully
mark the position of a camera with which he took a photograph of his
desktop. He then would invite the challenger into his office to make the
change behind closed doors.  After the alteration was made, he would
reposition the camera, and take a second photograph.  The before and
after photographs were mounted as a stereo pair, and the difference
would become immediately apparent the moment that he looked at them.
(see also "Responses of the primary visual cortical neurons to binocular
disparity without depth perception", B.G. Cumming & A.J. Parker, Nature,
Vol. 389 18 September 1997)

Your eyes are very sensitive to differences between them.

Back to the subject of the requirements for stereoscopic imaging...

One of the precognitive ways that the visual system get cues to
construct stereopsis is by the detection of (primarily) horizontal
differences in the images of the left and right eyes.  In fact, the
visual system is far more sensitive in the horizontal direction than in
the vertical; if it were not, television would look even worse than it
does. This is also one of the reasons why "standard" fax resolution is
200dpi in the horizontal direction, and only 100dpi in the vertical
direction.

So.... lets invert this question and think about how we could reduce the
cues required for stereopsis. Two that probably come to mind are:
1) blur the spatial differences and,
2) homogenize the chromatic differences.
And you would be right!  

An easy (and low cost way) to rid an image of spatial detail is to lower
its resolution. Referred to as digitizing or quantization error,
everyone is pretty familiar with this when the image on the computer
monitor does not look as good as the original photograph.  But remember,
that's only the error on one image.  When the binocular disparity would
have it's origins in detail that hides between the pixels scanned
images, then the disparity will be missing, or significantly reduced, by
the digitizing process.

Now lets look (he-he) at the more subtle characteristic of color
difference.  Sometimes, after quantization, the spatial cue gets
converted into a color difference (I could go into a long explanation
about why this is, but if you are still reading this, I don't want to
risk loosing you now).  And that's why JPEG may work great for
monoscopic images, but is a poor choice for electronic stereoscopic
images.  JPEG retains the spatial resolution of the original image, but
homogenizes the colors of adjacent pixels to achieve high data
compression ratios.

That is why, even though I don't always scan my stereo slides at
2,700DPI (unless I'm going to put them back out to a film recorder), I
concluded that it is important to scan stereoscopic slides at 36-bits
per pixel in order to retain the color differences.  Film scanners that
deliver 36-bit color resolution come with high spatial resolution.  The
40 second scan time is desirable if you scan a lot.  By the way, if one
is going to take the time to carefully scan a stereo slide, why not scan
it at a higher resolution that what you are using right now?  That way,
when the next generation of displays come along, then you won't have to
scan it all over again.

I don't think that the slide scanner referred to is overkill for
stereoscopic slides, even though it is way more than would be required
to put up a small 2D images on a web page.  That's not what I'm using it
for.

Jon Gross
http://www.frostbit.com/Catalyst/stereoscopic.html


------------------------------