Re: ScanROM 4E

[P3D Peter Davis <pd@xxxxxxxxxxxxx>]

At 06:38 PM 3/18/97 -0600, you wrote:
>P3D Peter F Davis wrote:
>> 
>> John Ohrt wrote:
>> 
>> > In an ideal world, your scanner needs twice the resolution of your
>> > printer.  So if you have an inkjet which requires 4 dots to compose a
>> > pixel, you divide the inkjets resolution by 2 in each dimension (ie
>> > 720x720 dpi -> 360x360 pixels per inch and then double each dimension to
>> > get the minimum scanner resolution, 720x720 pixels per inch).  If you
>> > have a dyesub printer, 1 dot -> 1 pixel, so if you have a 300x300 dpi
>> > dyesub, ideally you want a 600x600 pixel per inch scanner.
>> 
>> This is absolutely not true.  Each printer "dot" is either on or off,
>> so you only get two values per color.  The way halftoning works to get
>> intermediate color values is by clustering dots to approximate the
>> effect of larger dots.
>
>Nope.  The way halftoning works is to vary the dotsize.  Ask any
>printer.

But digital printers don't HAVE variable dot size.  What they do is use
multiple physical device "dots" to make one halftone dot, and vary the
"size" by controlling how many physical dots are on or off.

>> Consider the black and white case:  If you printed a 600dpi scan on a
>> 600dpi laser printer, you would only get pure blacks and whites ... no
>> grays.  However, if your printer resolution was twice the scan
>> resolution, you'd get effectively 4 gray levels, 4 times the
>> resolution gives you 16 gray levels, etc.  So your printer resolution
>> needs to be HIGHER than your scanner resolution, not the other way
>> around.
>
>Nope. You are confusing spatial resoltution with intensity resolution.
>
>You can simulate an enhanced intensity resolution, but the pixel is the
>area needed to encompus all those dots!  So if it takes 16 dots to give
>you sixteen grey scale levels and the printer is 600x600 dpi, it is
>150x150 pixels per inch, requiring a scan of 300 pixels per inch.

Well, you just confirmed what I said above ... the scanning resolution is
LOWER than the printer resolution, because it takes some number of printer
dots to create the effect of one variable intensity halftone dot.  Also, 16
levels of intensity is extremely small for any kind of photographic output.
You probably want something like 256 or more, which means that the halftone
screen resolution can only be 75 lines/inch for a a 600dpi printer.

>The Nyquist theorem relates solely to spatial resolution and roughly
>states that if you do not sample the input at an optical spatial
>resolution in excess of twice the final spatial resolution, then
>aliasing will occur.  The only way to prevent aliasing when you violate
>this theorem is to destroy information by filtering.  Sometimes though,
>the aliasing isn't noticable/objectionable, but that is a judgement
>call.

But, as we determined, the spatial resolution of the output device is
determined by the "screen density" rather than the device's physical
resolution, as we determined above.  For a 600dpi printer, using a 60lpi
(lines/inch) screen density will allow you 100 gray levels.

Nyquist doesn't really enter into it here, since we're talking about a
digital signal being output digitally.  Yes, the Nyquist theorem would help
you determine scanning resolution relative to the frequency in the original
photograph, but not relative to the output.  If you outputting at, say
75x75x256, you would sample at 75dpi.

>> Also, when resolution is specified for a color printer, are they
>> counting each color as a separate dot?
>
>For an inkjet or colour electrostatic, or a press yes!

Wrong!  A printing press doesn't have separate spaces for the different
colors.  The screens are rotated and placed on top of each other, independently.

>Because they are using a CYMK process requiring separate dots.

Yes, but it's not a grid with separate spaces for cyan, magenta, yellow and
black dots.  They all overlap.

>>  Typically,
>> the colors are layed down over each other, so they occupy the same
>> pixel space.  Over years of traditional printing, conventions have
>> evolved about rotating the screens (the grid of possible dots) for
>> each color to avoid moire patterns, etc.
>
>You are correct in your description of the technique, but a 400 lpi
>colour image only has 133 lpi spatial resolution on a press.

Could you explain this to me?  A 400 lpi halftone screen (higher than would
be used in any application I can think of in offset printing) requires 400
dots per inch.

>ps. I have worked on prepress work.  Also lpi is tricky to compare to
>pixels per inch because you have to understand the technologies and the
>intracies and assess the quality of each particular image.

I've worked in prepress as well.  LPI refers only to the density of the
halftone screen, which is then approximated on digital printers by some
combination of physical device pixels.

-pd
--------
                                Peter Davis
                     URL:  http://world.std.com/~pd/

                 "Nondescript -- the one word oxymoron."


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