T3D Re: digital camera resolution

[John Ohrt <johrt@xxxxxxx>]

John Bercovitz wrote:
> 
> > The Nyquist rules requires more than 2 pixels per line, to be precise.
> 
> Oops!  Can you tell us more about how that works?  I had heard
> that you had to have one column of pixels for the black line and
> an adjacent column for the white line.  I suppose that's only
> if you have perfect registration between the lines on the test
> target and the columns on the CCD.

Lets say you have precisely 75 lpm and scan at precisely 150 lpm.  The
resulting waveform is a straight line, ie 0 lpm.  Aliasing has occurred.

In a perfect world, scanning at 150.000000000001 lpm would solve the
problem.  In the real world, you have to be certain to get enough s/n
margin to ensure that aliasing does not occur.  Every case is different
and of course how do you really KNOW the data is at or below 75 lpm?

I am not suggesting that a casual user must analyze each image.  A
simple trick which is often good enough is to scan a max opical
resolution, or as high as the disk space free permits, and then
digitally resample by a method using integration.  You can fake this
process if your image processor doesn't truly integrate during
resampling by smearing the scanned output with a gaussian filter before
resampling.

To really KNOW, you want to inspect a 2D fft of the image and filter in
the frequency domain before using the inverse 2D fft to return to the
image domain.  I wouldn't worry unless you are doing scientific or
technical work.  There is software that does this, and NIH image or PC
image ( the latter is the Windows port of the NIH Mac package) are both
easy to use and free.  Easy is relative.

Remember the filter must be a "real" filter unless you are looking
forward to imaginary results  <grin>.

Regards,
--
John Ohrt,
Toronto * Ontario * Canada


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