Re: No, No, No

----------------------------------------------
If it is a linear sharpening filter, then it doesn't matter if it is
applied before or after the enlargement (with appropriate scaling).
Sorry, but the burden of proof lies on your
shoulders.
----------------------------------------------
Wrong. The burden of proof, if any, lies with the person making the
statement. You are the person making a statenment. I never said that it
could or couldn't be done so the burden of prove lies with you. However,
I said that linear operations are generally not commutative. And, don't
forget that the presharpening can also be used with many nonlinear
enlargement methods.

----------------------------------------------
So what? Well, that means your "scientifically based" method is
nothing new.
----------------------------------------------
I never said it was new. I said it was loosely based on a paper. I now
add that I suspect it is better and more general than previous
implementations.

----------------------------------------------
So your test image has, by definition, no noise ==> infinite PSNR.
Please explain how you can get a higher PSNR.
----------------------------------------------
Wrong. That is the original definition. It is now common convention to
use PSNR as a fidelity measure for signals with nonrandom errors as well
as signals with noise. Often deteministic errors are treated as noise. If
you don't already know this, then it is you who don't belong here.

----------------------------------------------
If the data-dependent Lanczos interpolator provides a higher PSNR, how
much does the pre-sharpening contribute to the higher PSNR for your
images?
----------------------------------------------
Answered in new Lena examples.

----------------------------------------------
Yes, PSNR is an objective measurement, but it does *not* imply "closer to
the true image."
----------------------------------------------
It implies the RMS error is smaller. There are other distance measures
such mean absolute error which, by coincidence, are also smaller with
presharpening.


.

Quantcast