|
PERSONALIZED 3D VIEWING A Brief Review John Hart Program in Atmospheric and Oceanic Sciences University of Colorado Boulder, CO 80302 nimbus.colorado.edu/hart/science.htm July 9, 2006
3D VIEWING OF DIGITAL SOURCE MATERIAL
Great 3D. What is it? Immersion. What is this? Some say it’s the art that matters. For example: Composition Depth of Field Net Deviation (net separation difference between near and far points, largely a compositional issue, though some technical aspects enter) Color and Contrast Some say it’s technical stuff that matters: Resolution of the Material Being Viewed (i.e. “Medium Format” is better the “Digital”) Viewing Method (lack of ghosts, color fidelity, lack of distortion, etc) Obviously, in the end, it is a combination of BOTH that matters. But let’s look at the latter. HAND VIEWERS So what is the ultimate VIEWING METHOD for looking at left-right pairs of stereo material? Some say 35mm transparencies (slides, actually usually 32mm wide by 24m tall) in a hand-viewer. Some go even further and insist on “medium format”, or “MF” slides (typically 50mm wide by 40mm to 50mm tall) in a hand-viewer. Reference to hand viewers: http://www.berezin.com/3d/viewer_selection_chart.htm Hand-viewers are small, compact, and easily used. Some focus well and have interocular adjustments. Some don’t do either, and can be impossible to use, depending on your vision. These devices usually offer a pair of reasonably corrected lenses that image individual pairs of slides. There is little cross-talk between the images (so the right eye sees only the right image and there is no “ghosting”). Uniformity of lighting, color, and contrast can be quite good. But, there are several raps against hand-viewers. You have to mount the slides carefully to avoid registration errors, perhaps not as carefully as for projection, but you should do this anyway. It’s a one person sort of game, of course, where the material can’t be viewed, even obliquely, by another soul. This can limit discussion and interaction. The “material” can’t easily be distributed to the public (e.g. to all those having a hand-viewer), simultaneously. Of course this can be done, but only by the expensive prospect of copying (or “duping”) slides. Duped slides, even at $3+ each, are fuzzy shadows of the originals when viewed at the moderately high magnification typical of most slide hand-viewers. http://www.crystalcanyons.net/Pages/TechNotes/FilmVsDupes.shtm HAND VIEWERS WITH DIGITAL SOURCES? There is another issue. Hand-viewing is not readily accessible to those with digital cameras. The only way to use a hand-viewer is to take pictures of your computer monitor or “film-record” your digital files to transparencies, using a machine that is basically a CRT flying spot scanner that transfers 3 colors (via 3 passes with RGB filters) to film. Typical 35mm film-recorders produce “slides” that are about as good as slide dupes. References: http://www.crystalcanyons.net/Pages/TechNotes/FilmRecorderTest.shtm http://www.crystalcanyons.net/Pages/TechNotes/ShootYourMonitor.shtm Commercial labs (like http://www.slides.com ) produce pin-registered output (which circumvents the individual mounting of images into slide mounts), but charge ~$5 per pair.
I have recently tested medium format (MF) transparency output from 5000x4000 pixel digital files. These slides were made by slides.com on their elegant Solitaire Cine3 recorder. This device puts images onto 70mm film. It is thought to be among the “best” available. It is being used in Hollywood to produce IMAX films and was used to master Superman in IMAX format. The cost ranges from $5 per 50mm x 37mm (4:3 aspect ratio MF) pair, to $10 per 60mm x 70mm pair (for which there is no useful hand-viewer). Are these as good as original images shot on fine-grain film with a real medium format camera? Probably not, if viewed under a microscope. However, you don’t see the granulation fog that 35mm film-recorder output exhibits when observed under magnification. In addition, the PERCEIVED RESOLUTION, i.e. the resolution that the human eye can actually resolve in a specific viewer, is probably close to that obtained with an original Velvia MF slide (see below). Which is better? The MF film-recorder output significantly exceeds the quality of “home” film-recorders like the Polaroid ProPallette 8000, as well typical commercial 35mm film recorders like the MGI Sapphire. Although such machines claim 4000 or even 8000 pixel resolution across the image, blooming and finite spot size SIGNIFICANTLY reduce the actual image resolution (see the above references to Shoot Your Monitor and FilmRecorderTests). I did not see much difference between the home-brew PP8000 and Slides.com’s pin-registered MGI Sapphire. MF film-recorder output in a MF hand-viewer is quite good. Regardless of cost, Is it the best way to view material acquired with digital cameras? Surprise! NO in is NOT!!
VIEWING RATIOS Before presenting what I consider the very best 3D viewing method currently available for digital-source material (i.e. digital camera output files, or digital scans of film), it is useful to summarize some VIEWING RATIO and RESOLUTION issues. An important quantity in 3D viewing is the VIEWING RATIO. This is defined as the distance at which the stereo pair is observed, divided by the image width.
VR = (viewing distance) / (image width) (1)
VR’s approaching 1 tend to be “IMMERSIVE”, though this magical and somewhat mysterious quantity is also a function of composition, lighting, resolution, etc. Now, it’s easy to estimate VR when viewing a print with an un-aided (no magnifying glasses) human eye. Just measure the distance from the eye to the print, and divide by the print-width. For me, as it gets down close to 1, it gets immersive. What about the VR for hand viewers or other LENS-assisted systems? If the lenses are set at infinity, then the VR for a hand-viewer or lens-based print viewer is just the lens focal length divided by the image width. If the viewer lenses are set at some intermediate image-distance, VR goes down some, but for ease of use by most people, the lenses should be set to image at something like 8 feet or more so both near-sighted and far-sighted people can use it. In this case the VR of lens based viewers is, with small error,
VR = focal-length/image width. (2)
For a mirror based viewer, like David Lee’s Hyperview or the View-Magic double periscope devices, just track the light ray from the image, bouncing off each mirror, to the eye (thus the viewing distance is longer than the straight distance from the viewer to the image. References: http://www.berezin.com/3d/ViewMagic.htm http://www.berezin.com/3d/Hyperview.htm http://www.skep.com/3D/gallery.htm Here are the VR’s (rounded up to nearest tenth) for a number of viewers: Combi 35mm slide viewer 1.6 StealtheLight 35mm slide viewer 1.3 CO5 35mm slide viewer 1.9 Saturn (Alan Lewis) Medium Format slide viewer 1.6 Sam Smith’s Regal MF viewer 1.6 Sam Smith’s MF Woody Viewer 2.0 View Magic Print Viewer (4” and 8”) 2.0 Hyperview Print Viewer 2.0 Holmes Standard Viewer (typical) ~2
Thus, apart from the StealTheLight 35mm slide viewer, all these units have VR > 1.6 . This means less immersion in the image, and has consequences for observable resolution.
Note-1: Anaglyph prints and Stereojet prints can be observed at VR ~ 1, even less. However color accuracy and contrast/saturation can be problematic for these media (and Stereojets are no longer commercially available) so they are not considered here. Note-2: How is VR related to the so-called Ortho-Seat, where a stereo view will look the same, with respect to perspective, as in the original scene? They are independent. However if you want an ortho-perspective, you have to sit at the ortho-distance. The ortho-distance is OD = (image width)*(focal length of taking lens)/(sensor/film width) (3) For example, for a medium format (MF) film (film width ~50mm) camera taking shots with an 80mm lens, the OD for viewing a 50mm wide slide is just 80mm. For viewing a 100mm wide print it’s 16cm. From (1) and (3) you can see that for a full frame image produced with a lens having a focal length equal to the film width, then the viewing distance for VR = 1 is exactly OD. Such lenses are usually categorized as moderately “wide angle”. Wide angle scenes tend to be more “immersive”, though that’s perhaps treading on composition issues. In any case VR ~ 1 should be strived for when images are taken at modest wide angle.
In Summary, few current stereo viewers have immersive, per se, VR. Unless you look at MF with a VR of 1 (which nobody does), we can do just as well, detail wise, with digital prints, as I now show. RESOLUTION ISSUES This area is most controversial, so I won’t belabor it. In quick summary (see references for more in-depth discussion): 1) 20-20 vision is by definition the capability of resolving an object 1 arc-minute (1/60 of a degree) wide. 2) This corresponds to 30 line pairs per degree, or 30 cycles per degree (cpd). 3) The real eye (like film and digital) has contrast roll-off. Vision just doesn’t cut off at this 20-20 limit, but objects start to appear more and more fuzzy at 10 cpd and beyond For example, at 10cpd, the eye is already down to 50% contrast. Ref: http://www.normankoren.com/Tutorials/MTF.html 4) A Film-Camera system behaves similarly. You don’t get 100% contrast at 50 lines per mm. Now, if you really needed a digital system to resolve 50 lpmm at ~100 percent contrast (using 2 pixels per line), you would need 24*100*2*36*100*2 ~ 30Megapixels to match a full frame of 35mm film. But digital cameras can produce results as impressive as film even with a relatively smaller number of pixels. Why? If the 50 lpmm on film are only at 50% contrast (as is typical), then you only need 1 pixel per line instead of 2 (e.g. Synchronous lines give you 100% contrast, totally asynchronous gives you 0%, This averages to 50% - how’s that for math ;-). Then you can match 35mm film with 30/4 ~ 8 megapixels. There are also issues with film grain and clumps (see the third reference below for some striking photomicrographs). References: http://www.luminous-landscape.com/essays/clumps.shtml http://www.normankoren.com/Tutorials/MTF7.html http://www.crystalcanyons.net/Pages/TechNotes/FilmVsDupes.shtm In summary, a 10 or 12 megapixel digital camera can produce PRINTS that roughly equal those produced by scanning 35mm film. A 17 megapixel digital camera can produce prints that for many shots roughly equal those produced by scanning MF film with the best drum scanners (or those produced in the wet darkroom).
PUT IT TOGETHER Now here’s the main point! At a VR of 1.6 to 2, the detail the human eye can see in an image drops by exactly these factors. The total resolution you need in the image, to be “rezzed-out” – to introduce a new saying, goes down by the SQUARE of these numbers. Example: For a VR = 2 viewer you only need about 4 megapixels (17 / (2*2) ) to produce a print that in principle contains most all that can be seen. Note – 4 megapixels in each RGB color corresponds to about an 8 megapixel Bayer-grid sensor.
Thus, WE SHOULD BE ABLE TO PROVIDE AS GOOD AN EXPERIENCE AS THAT SERVED UP BY CURRENT 35mm, MF, AND AVAILABLE PRINT-VIEWERS USING ~10 MEGAPIXEL SOURCE MATERIAL. BUT, the images have to be presented in as clean a way as possible, with no losses in going from digital camera files to the image to be viewed. Film recorders don’t do very well, as discussed above. This suggests that digital images should be presented, in hard copy, as PRINTS. The advantage of prints is that can be a totally controlled home enterprise. No labs. No uncorrectable processing errors like scratches, or worse. An Epson R800 ($350, vs. $thousands for a film-recorder) makes spectacular prints, up to 8.5 by 11. Alternatively, one could use high-quality commercial printing to distribute books. The R800 prints are good enough to outshine MF film recorder output (by my tests), and probably good enough to roughly equal MF originals in a VR = 2 hand viewer. The Epson R800 can lay down about 6 lp/mm (150 lp/inch) with very good contrast. On a 10” wide print that’s about 1500 line pairs across the image. This is about what the human eye can resolve, at a VR of 1 !! At a VR of 1.6 or 2, such prints will look JUST AS GOOD AS MF, because at this VR both the prints and the MF will exceed the resolving power of the human eye. That is, at VR = 2 the eye can resolve about 800 lp across the image, but you’ve got 1500 or so. Ref: http://www.crystalcanyons.net/Pages/TechNotes/R800Printer.shtm
|
