Re: digital/analog

[P3D John W Roberts <roberts@xxxxxxxxxxxxxxxxx>]

>Date: Sat, 14 Dec 1996 20:52:46 -0600
>From: P3D john bercovitz  <bercov@xxxxxxxxxxx>
>Subject: digital/analog

>Here's something to chew on that has to do with digital.
 
>Medium format stereo has an angle of view which is 36 degrees wide 
>and 36 degrees high.  7P format has an angle of view which is 34 
>degrees wide and 29 degrees high.  5P has an angle of view which 
>is 27 degrees wide and 29 degrees high.  V-M has an angle of view 
>which is 13 degrees wide and 12 degrees high.
 
>The human visual system can resolve about 1 minute of arc and can 
>use its rods and cones in some sort of algorithm which gives it a 
>stereo or vernier acuity of 10 seconds of arc.  I don't know how 
>many colors the eye can discriminate but I bet you computer 
>graphics folks can tell us in an instant.

More than 24 bits per pixel (8 bits = 256 grayscale levels per color)
is fairly unusual, in part because it's extremely difficult to get a
display system to show even that many brightness levels. A Kodak paper
recommends 10 bits per color out of a dynamic range of 16 bits per color
(30 bits per pixel, plus whatever is used to handle the range information).
Some people reportedly like 12 bits per color per pixel. The human eye can
see a dynamic range of about 1,000,000:1, so if you want to represent that
as RGB values, that would be ~60 bits/pixel. It would be immensely difficult
to either gather or display this great a range using any medium, including
film.

A few comments:
 - Human vision handles brightness and color somewhat independently - if
     information is stored with this in mind, a lot of bits can be saved.
 - In general the chroma information can be stored at lower resolution than
     the luminance information, with little or no loss in perceived image
     quality. Blue resolution is particularly low.
 - Established image compression methods can help considerably in reducing
     required storage. 2:1 compression is often possible with no loss.
     Lossy compression at ~10:1 can often look very good. Some methods
     can get 50:1 to 100:1 lossy compression with surprisingly good results
     considering the amount of information that's thrown away.
 - A note of caution on lossy compression: For a given image and compression
     method, as the compression ratio is increased, the reproduced image
     eventually degrades, typically by the appearance of artifacts (reduced
     resolution, "echoes" of edges, etc.). Most of the work in image
     compression seems to be directed toward keeping these artifacts below
     a threshold that is acceptable for 2D viewing. But 3D viewing is usually
     much more sensitive to artifacts, especially if the artifacts don't
     match in the left and right images. I believe that investigation into
     this issue would be very beneficial in the long term for 3D photography
     and imaging.
 
>Given these two inputs, how much data would be required to 
>simulate digitally any of the stereo formats we have now with full 
>stereo resolution?  That is, what would the file size be for a 
>single picture?
 
Tough question! I think specifying it by angular dimensions and the total
range of human visual capability would be to describe a system vastly better
(possibly thousands of times better) than any actual existing viewer and
stereo photograph systems. For more realistic numbers, we need more
information on the optics (lens flare and such) of the cameras and
viewers, and the properties of the film typically used in traditional
stereo cameras (grain size, color reproduction, total dynamic range, etc.).
One thing in particular that confused me this week was the posting of two
confident assertions:
 (1) the individual grains on the film are capable of recording an
       (effectively) infinite number of brightness levels
 (2) the individual grains on the film are essentially either "on" or "off".

Given the level of uncertainty indicated by disagreement on such fundamental
issues, I think we might have a lot of trouble figuring out what the
traditional film, cameras, and viewers are actually capable of, and
therefore of what it would take to reproduce that capability with an
all-digital system.

[Quick and dirty answer: find out the resolution and bits per pixel
represented in the highest resolution image in a 35mm Photo-CD file,
calculate out "bits per square millimeter" on a 35mm negative, and
scale that to the various stereo film chip sizes. (Then multiply by two
for the left and right views! :-)  The actual amount of information needed
will probably be less than what you calculated, and maybe a lot less.
Also bear in mind that one of the ways to compress stereo images that is
being developed is to replace one of the views with a "difference file"
relative to the other view - for most stereo pairs that gives nearly 2:1
compression virtually free.]

BTW: My number for human vernier (not stereo) acuity is *one* second of
arc (reference: the human factors people at Xerox Parc, 1993).

John R


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