Re: Scientifically Based Presharpening for Enlargement



Dave Martindale wrote:


I don't understand this business about halos. How does the presence of
halos indicate "presharpening is not simply compensating for loss due to
finite sensel area". The fact that the sensor has area means that all
frequencies that are greater than DC will be attenuated when the image
is captured. To compensate for this all frequencies greater than DC need
to be amplified. This is what a sharpening type filter does for all
frequencies except at (or very near) Fs/2. I don't see what is the
problem with that?

If the sharpening used merely compensated for the attenuation during
image capture, it would give an image that was closer to the original
scene before those losses took effect.

No, it would do that only if the original had been bandlimited. That
could have been done. I gather from what the OP has said no effort was
made to band limit the original because the object is to simulate the
capture of an image that is not band-limited and to avoid a blurry
reconstruction.
Remember if you reconstruct the low-res image perfectly at 4x
enlargement the result will have no frequency content above fs/8. That
will be pretty blurry. So if the starting premise is to avoid a blurry
result, you have by definition from the start precluded classic
reconstruction.


So the "ideal" sharpening
filter, when used for this purpose, would have a frequency response
that is the reciprocal of the attenuation during capture.

Thr halo's are not evidence that this is not the case as you are
claiming. The halo's are consistent with the OP's claim that an image
(that was not bandlimited to the lower resolution) was downsampled by a
box-car filter by a factor of 4 and a sharpening filter applied. As it
happens a sharpening filter can be pretty close to having the inverse
response to the box-car filter. But I don't know if that was the goal.

If the intent was to find the ideal reciprocal filter then the original
image should be first be filtered to remove all frequency content above
fs/8 so that aliasing does not occur when it is downsampled by 4. Then
the capability of the filter could be accuarately measured without the
interference from aliasing. But apparently, the intent is to include and
work around the aliasing and not to avoid it.

The aim is
to get an overall response, after sampling and compensation, that is as
close to flat as possible (below Nyquist).

Well, If I understood correctly the aim was to simulate the capture of a
non-bandlimited image and acheive the closest reconstruction possible as
defined by the several objective methods used to measure the difference.
The only problem I see is that what is measured to be the best is not
necessarily what most people are going to want to see.


-jim


But in this example, there are halos around dark/light transitions that
were not present in the original scene. This indicates that the
amplitude of some spatial frequencies has been boosted above what it
was in the original, and so the sharpening has gone beyond restoration
into enhancement. It's not just compensating for attenuation during
sampling.

Dave

----== Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
----= East and West-Coast Server Farms - Total Privacy via Encryption =----
.

Relevant Pages

  • Re: Earl & Donna Evleths address
    ... >> Spam filters are quite good, our Wanadoo filter gets about 90% of it ... But they filter most of the stuff without ... I can switch frequencies and go back 50 years in behavior, ...
    (rec.travel.europe)
  • Linear phase eq using overlap add problem
    ... Incoming blocks are then added like this: ... a 100Hz sine and used some +24dB high-Q lowpass, ... suspected that it takes some time for the filter to ... creating crackles and also frequencies that really were not in the signal! ...
    (comp.dsp)
  • Re: Amplitude response of a Biquad Filter
    ... Wrong implementation of filter coefficients???? ... But, when I use an ordinary audio file, I see the ... Sampling frequencies ranging from 8 kHz to 48 kHz (audio related ...
    (comp.dsp)
  • Re: Sampled Data and Frequency Rejection RC Filter Question
    ... We don't have a big-budget filter. ... How big should our RC time constant be? ... I don't see that the frequencies in [500Hz, 1000Hz) will do us any harm. ... We aren't trying to reconstruct signals. ...
    (comp.arch.embedded)
  • Re: method for digital filtering
    ... not saying the sampling rate increases as a function of the load ... Is it the weight of the object or the speed ... Why do I get the IR If I high pass filter the signal? ... Yes and one might guess that the low frequencies contain the information ...
    (comp.dsp)