Re: 8 verses 16 bit

[Kennedy McEwen <rkm@xxxxxxxxxxxxxxxxxxxx>]

In article <OxYCoLLi7Ah8Ewii@kennedym.demon.co.uk>, Kennedy McEwen 
<rkm@kennedym.demon.co.uk> writes
>I asked yesterday but you haven't produced any info.  Similarly with 
>the "human vision studies" that you cited.
>
Part II:
In article <Pine.GSO.4.44.0203022005350.28181-100000@e4e.oac.uci.edu>, 
David Chien <chiendh@uci.edu> writes

>As a result, using a 8-bit color system immediately prevents you from
>recording subtle color shading in certain types of photography - eg.
>flowers, skin tones, etc.
>
That is true - it prevents you from recording them, but not from SEEING 
them - your own visual system imposes that limitation.  In order to see 
them and differentiate them one from another you need to modify the 
levels in the image significantly in that local region of the palette. 
Even in the limited 14-bit range that most of the better scanners 
operate in, the extra 6 bits per channel only serve as a "safety margin" 
to accommodate imperfect black and white point setting of either the 
operator or the scanner designer.

8bit scanning requires(d) that you determine black/white points and 
gamma with high precision since this range is only slightly greater than 
the range of colours that the eye can discern.  Further processing of 
either of these parameters by necessity introduces visible 
discontinuities.  However, when set correctly in the first place, 8 bits 
is more than up to the job of image reproduction.
>
>Now, what happens when you manipulate these colors in a photo editing
>program?  If you increase the contrast/brightness/gamma/etc. of the image
>what you will notice is 'banding', or sudden jumps in what was otherwise
>smoothly shaded areas of color, casued by the lack of intermidiate colors
>that are required for accurate color adjustments.
>
>You can understand it this way, if you only have two shades of red,
>very bright red, and very dark red, how do you display a blush red?
>
>You can't because there isn't enough colors in between - this is what
>an 8-bit color system does to your images before and after editing.
>
Only after editing - NOT before!  ;-)
>
>So how many more colors are enough?
>
>One simply looks to various sources - film scanners, 3D raytraycing used
>in movies, human vision studies, 35mm film, etc. and you'll see a varying
>range being used, all greater than 8-bits because it has long been known
>to be inadaquate.
>
>16-64 bits are all in use

Name one scanner which uses 64bits per colour!  For that matter, name 
one that uses more than 16 bits at the ADC!

You seem to be losing the distinction between the number of bits PER 
COLOUR and the number of bits in ALL COLOURS.  Certainly there are 
scanners which fall into the latter category at 48 bits, but none that 
meet a 64 bit level - yet you are suggesting that there are some in the 
former category as well.

Name them!

>, and I myself would prefer at least a 32-bit
>color system to stay in sync with higher-bit applications in other
>industries (esp. 3D animation for movies and commercials), to be
>able to display that 32-bit image w/o any dithering or color conversion
>at all on most 32-bit capable graphics cards (just about every video
>card made today can be set to run in 32-bit mode =- we're not talking
>about cheapies cards here, either.  At least a GeForce!)
>
That is losing the plot David.  The Geoforce cards output 8-bit colour 
in all of their modes - 16 million colours.  8 bits in each colour 
represents 24 bits total. The remaining 8 bits are used for texture 
overlays, lighting, antialiasing and 3-D geometry effects and 
acceleration functions.  You don't get 32 bits of colour information 
output, just 24 - 8 bits for each colour!  You appear to be confusing 
32-bit processing of the nVidia chipsets with their output image bit 
depth.  Some of the better cards use 10-bit DACs - but only provide 
8-bit data to them, the higher resolution DAC being used to ensure a 
"real" 8-bit linearity figure.
>
>A quick look at the histogram in any paint program (preferably Paint Shop
>Pro because Photoshop 1-6, and probably 7, still STUPIDLY does not allow
>you to have it up all the time to let you color correct in real-time by
>eyeballing it) will let you see the harsh effects of an 8-bit color
>system.
>
>Simply take any image, then push the gamma/brightness, contrast, hue,
>saturation, up a lot.  You'll notice that the histogram oftenbecomes
>broken, with a smooth curve turned into a curve that has a lot of thin
>slices taken out of it.
>
>This is what happens when the system does not have enough intermediate
>colors to dither to.
>
>While this still occurs in a 16+bit system, the visual artifacts such as
>banding are far, far less noticable, if at all.  More bits simply lets you
>calculate things far more accurately.
>
This is NOT giving you more colours to see - it is manipulating the data 
that you have captured INTO the range of colours that you can see: a 
range which is completely encompassed by 8-bit representation of the 
primary colours under an appropriate gamma curve.

The same effect would ultimately be apparent if you could get an 
instrument to resolve a thousand bits in each colour - giving a total of
1.230231922161117176931558813277 x 10^903 different colours.  Since this 
is larger than any estimate of the total number of particles in the 
known universe by 820 orders of magnitude, I am fairly certain that the 
eye cannot see this number of discrete colours.  Hence, since this also 
fails your test of adequacy, I suggest that the argument you have 
presented is flawed, not the choice of the number of bits displayed! :-)
>
>Also, 16-32bit systems would let everyone work with their raw scans and
>digicam images without further dithering and translation into a lower
>8-bit image, which also introduces inaccuracies and throws away good data.
>
>Film scanners often work in 10-16bit modes, and high-end digicams from
>such makers as BetterLight.com and PhaseOne.com (yep, $10,000+ 5000x5000
>to 10,000x10,000 pixel digital cameras!) can output raw 16+bit images.
>
So why do you want 32 bits per colour if no instrument used in the 
photographic reproduction process can utilise it?  Even those you have 
presented are instruments which are capable of finer colour 
discrimination than the eye can discern.
>
>Aside from the commerical uses, the medical industry would also benefit
>from photo editors that use 16-64bit image modes.
>
>After all, you definitely don't want a spot to be misdiagnosed as cancer,
>let's say, would you?
>
Come on David, you can do better than tug at the old bleating 
heartstrings - that is just plain BS: and you KNOW it!

>Many scientific and other industries would benefit greatly from a
>higher-bit photo editor.
>
Why?  You have yet to identify ANY imaging instrument which produces 
more than 16-bits per colour.  Some use 8, most seem to be in the 
10-14bit range and a few use 16 bits - NONE use 64 bits per colour 
(which I assume to be the measure you to be referring to since PS, 
operating at 8bits per colour, already is a 24-bit photo editor in all 
it functions and a 48-bit editor in many).
>
>   Anyways, all this talk may or may not be useful for the layman who
>rarely color calibrates, color manages, or cares for the 'most' accurate
>colors his PC system can display today.  He may simply ignore all of our
>talk, sit back and enjoy the 8-bit age for the rest of his life.
>
>   For the rest of us neurotic types, I say it's rather 'lame' excuse that
>after 6+ generations of Photoshop, not one thing has been done to migrate
>the entire system to a flexible non-hardcoded-bit based system!
>
Hardly surprising if the best argument you can come up with is "it would 
be useful for imaging sensors that don't exist yet and might never do in 
the future"!

Whilst there IS an argument for Photoshop evolving to n-bits per channel 
for all operations (the counter argument of insisting that certain 
operations are implemented in a specific order - maintaining full source 
dynamic range until output eliminates such restrictions) it is certainly 
difficult to argue that n should be greater than 16 - and the argument 
that you have presented is so full of inconsistencies that it does 
neither.
-- 
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers
-
Turn off HTML mail features. Keep quoted material short. Use accurate
subject lines. http://www.leben.com/lists for list instructions.