P3D Re: Request for feedback on cha-cha's

[Larry Berlin <lberlin@xxxxxxxxx>]

>Date: Thu, 18 Mar 1999
>From: Michael Kersenbrock <michaelk@xxxxxxxxxxx>
>.................
>Greg,
>	I'm curious too!  I'd appreciate feedback from folk
>        looking at:
>
>	    http://www.photo-3d.com/Csc/cscfeb99.html
>
>There are several cha-cha shots there.  Questions I have are:
>
>1) Do you have significant problems fusing them?
>   a) with the parallel view pair?
>   b) with the cross-eyed view pair?

******  No. But I missed having JPS versions... ;-)



>
>2) On your video screen, how wide is
>   a single image -- actually measured ?

*****  For *omsi_o.jpg that's 2 and 9/16ths inches. Using 1024 x 768 desktop.


>
>3) Can you detect rotation or keystoning problems?

*****   Some. I'm going to look in more detail to see what it is my eye
picked out. In some cases it's motion of people or cars.

In *caldepth.jpg* I see vertical misalignment with the Left eye's view
significantly higher than the Right eye's view. On closer examination it is
an image scale mismatch. The book in the Left eye is much closer to the top
of the image than in the Right eye's view. However the bottom of the images
appear to be perfectly or close to vertical alignment. Likely cause? Not
moving perpendicular to the center of the view while doing the cha cha. Left
eye is closer to objects than the right eye.

It has a stronger effect in triplets like this than it would with a single
pair in either format. The book in the image has an area (top right corner)
with strong retinal rivalry due to what looks like a stain on the book
jacket. It's quite different from each eye's view and is likely a reflection
problem. This is something that if noticed can easily be fixed in Photoshop
using the clone tool.

If it has further distortion problems I can't identify them from casual
observation. I have to fix the vertical alignment first since that might
account for anything my eyes pick up now.

If you are concerned about the apparent size of the images under a variety
of viewing circumstances, use the Stereoscope Applet. Viewers can choose to
view at percentages of file size from smaller to larger preset values and in
increments of 10%. As well as choosing parallel, crossed, or a range of
anaglyphs for method... You wouldn't need triplets as it works with stereo
pair image files or separate left and rights. You can specify one of the
many viewing modes available as the default first view. Thereafter it stays
on the last used mode selected by the user. Very convenient.

On the voting boxes...

My eyes pick out some problem. On examination, the Left eye's image scale is
enlarged compared to the Right eye's image. Whether this is due to the
camera moving closer to the object, or some other change I would have to
guess and make some measurements. If it's simply a scale problem, you would
identify the limits of common image content and trim away any excess. Then
shrink the largest image to the dimensions of the smaller one. If no other
problems exist it would become more viewable and each detail would be
aligned vertically.

On *cake.jpg*...

This image has the same image scale problem as the voting boxes. The Left
eye's view is enlarged. A mismatch of this type creates an uneven vertical
disparity to the viewer depending on what part of the image is being
observed. If the image is taken as a whole, there is no viewing position in
which all parts are equally aligned. In this case you would start at the top
with your head very very slightly tipped left. As you scan downwards you
slowly tilt your head towards the right. Part of the center is perfectly
aligned. At the bottom your head would be tilted to the right.

While this doesn't overly bother my experience or that of many others, it
could be the source of stress some with sensitive eyes will notice. The
amount of this error is not untypical among other stereo views in general
(unknown methods - generalized comparison).

Another way of noticing these comparisons is to make an anaglyph with them
(viewed without colored glasses). In the anaglyph, each corresponding detail
has it's own position in one color or the other. Where there is parallax
(horizontal separation), you can draw a line joining the two points. If this
line is not horizontal, there is a misalignment. Always look at several
points and compare top vs. bottom, and side to side. The result will be a
map for image correction.

An alternate way to make this same comparison is simply to blend the two
images at 50%... You don't need to be in anaglyph mode, you just need both
images visible in an overlapping comparitive space. You use the information
gained from comparing these copies to make corrections in the original
images. ( It even works looking at page flipped images without LCS glasses.
This is a case where a slower flicker rate works better than a fast one.
It's a good reason to install page flipping software even if you don't have
LCS glasses. It would work in interlaced mode too, but with significantly
less accuracy)


>....................
>What I'm asking is feedback on the basic cha-cha and 
>"mounting" when done in this size and freeviewed.  These
>particular images are expressly done as snap-shot
>recording of club activities for those who missed a 
>meeting and/or others who might be interested -- so I'm
>not looking so much for content issues (and those pages
>so far have been hit quite a few times -- thanks!).

******  I'm interested in the comparative experiences of side by side images
with and without a separating border. In a sense, a border helps the eyes to
lock onto the image. A border method can be applied to pairs or triplets or
any viewing method. It does increase the file size some, but if it adds to
the experience it might be worth it. Any comments from other's experiences
with/without image borders (mostly relative to freeviewing)?


>
>In terms of the previous thread, are these good enough
>(cha-cha's) for somebody to start with?  Not talking
>about stereoscopic perfection, but more along the lines
>of whether one can snap easy-ones and enjoy the results.

*****  Absolutely yes!


>...............
>I've been thinking of using larger cross-viewed-only
>to reduce retinal rivalry due to the low resolution and
>heavily compressed jpeg stereo images used, but 
>filesizes already are pretty big and I'm already
>compressing the heck out of them with very high
>compression levels (not to speak about my cheap 
>camera's original jpeg compression).

*****  General strategy in compression... Shoot with an absolute minimum of
wasted space around your subject of interest. Then trim as much excess as
possible. The result is an optimum resolution per subject. Finally, if
possible resample to some smaller size (than the original) for display
purposes. Save any edited work in a NO-loss format and make JPG copies from
there as needed. You can't help the source JPG in this case, but you don't
want to magnify it's problems more than necessary.

By going with the Stereoscope Applet you would be able to optimize the
balance between file sizes and compression ratios. Since you woudn't have
actual triplets, the surface area (now) for three images can contain two,
resulting in an increase of resolution per similar file size.

Judging by the fact that on pictures that I could find problems, they were
all the same problem, I would suggest that it's related to something
consistent in your method. That's good. Which side do you shoot first? I'll
bet you usually do it the same way! Practice with this pattern and use
physical references to train your sense of what is perpendicular as well as
level. That one new skill would fix most of your future images.

Keep up the Cha-cha!

Later,
Larry Berlin

Email: lberlin@xxxxxxxxx
http://www.sonic.net/~lberlin/
http://3dzine.simplenet.com/


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

End of PHOTO-3D Digest 3255
***************************