In a message dated 4/30/06 10:49:48 AM, hogarth@... writes:
> How can this be? My understanding of a pixel, any pixel, from any
> digital camera or film scanner, is that it's generally a square and can
> have exactly one RGB value, at some number of bits between 8 and 16
> (depending on the ADC), usually padded out to fill the 16 bit data words
> used by computers. That is, it can have at most 16 bits of R, 16 bits of
> G, and 16 bits of B. If you reduce to grayscale, you get at most 16 bits
> of luminance value only.
>
You are dealing with the math representing the value, not the meaningfulness
of the value. I suspect you would agree that a 5 megapixel image shot with a
low cost point and shoot camera with a tiny plastic lens and a tiny sensor
gives you less meaningful data than a 5 megapixel capture shot with a pro camera
with a larger sensor, and a large, expensive lens; it will not resolve line
pairs at as fine a level. So the content of all pixels are not created equal,
even if the pixel value, mathmatically, is. Same for scans. I had one client
supply me with a series of pro scans that had been done for him. They were lovely,
dreamlike content, but so fuzzy and soft in the original version that
scanning them at high rez was a waste of money, he could have extracted all the
meaningful data from them with a low cost flatbed. The rest was pixels full of
noise, redundancy, or at best film grain... maybe the grain could be helpful as a
level of pattern in blowups, but the image data was low rez in a highrez
format.
Scans often capture a whole lot of nothing. I recall scanning a series of
35mm slides shot on safari in Africa for a friend. I ran them en-mass on my large
format flatbed at its max DPI to make comps of the files, so we could decide
which ones to rescan with a dedicated film scanner. It proved unnecessary, as
the flatbed was already scanning at a higher rez than any of the images
needed, and film scans would have just been bigger files without any more meaningful
image data. That is to say; the lion's mane would not have been sharper,
clearer, less posterized, or in any visual way more detailed or meaningful with a
much higher rez scan... there would just have been a much larger file to deal
with. Short of getting into theoretical issues of how much grain detail is
enough grain detail, that happens sooner or later with all 35mm shots, and almost
always at a point below the meaningful resolution of the current digital
SLRs.
>
> Besides the size of the square and the RGB value what "more meaningful
> data" are you talking about?
>
> > High rez scans from small film are all about one level or
> > another of grain definition... and while there can be ISO noise in a
> > high rez
> > digital photo, there is significantly more info per pixel, especially
> > when shot with
> > high end cameras, high end lenses, and some user expertise.
>
> My understanding is that it's impossible to actually image film grain
> (that is, grain clumps or the resulting dye clouds in color films) with
> a scanner. The grains when seen under a microscope are generally
> fractals. The fractal detail (actual shapes) is considerably smaller
> than the 3 micron size of the best aperture on the best drum scanners. A
> 3 micron square is at best a poor representation of most film grain.
>
Anyone who as ever scanned film will tell you that it is easy enough to scan
filmgrain... I don't know where your theory comes from, but the empirical
results are quite different...
>
> From a film scanner then you get what I call "digital grain" which is
> not at all the same as film grain. It's a different kind of information,
> largely due to making a deterministic sampling of a stochastic media.
>
This is that endless discussion of how much data from film grain is enough
that I wasn't much interested in having. When you first reach the level of
seeing variations from grain, it forms a pattern that could well be called digital
grain: it varies, but it does not define the grain very well, rather like
determining with the tines of a metal rake that there are variations in the
smoothness of the surface of your garden, but the tines are too far apart to
actually define the pattern that they detect. More detail, higher rez scans go
further and further into defining the grain. Eventually most of the pixels in your
scan are about grain detail, not variations of any other kind. Since, on one
level, all data in the film is about filmgrain, you could claim this is all
relevant, and meaningful data, but any info about the image you shot, the ability
to resolve image detail, not film grain, ended long before. So, I leave you to
whatever degree of film grain fanaticism you choose to support, its all the
same to me, the only grain I have to deal with these days is in high ISO
digital camera shots.
> Very similar to what a digital camera does - making a deterministic
> sampling of the stochastic scene being transmitted to the sensor by the
> lens.
>
> Not when you talk about resolving lines per width of frame, not texture
> that was not part of the original scene, only the film shot of it...
>
> So again I'm curious as to what this "significantly more info per pixel"
> is. I don't understand it, please help me see what you are talking about.
>
> See above...
C. David Tobie
Product Technology Manager
ColorVision Business Division
DataColor Inc.
CDTobie@...
www.colorvision.com
[Non-text portions of this message have been removed]