[opendtv] Re: Math of oversampling - a simple comparison
- From: Jeroen Stessen <jeroen.stessen@xxxxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Wed, 27 Apr 2005 09:34:45 +0200
Hello,
Tom Barry wrote:
>> Okay, here's one from Leno. I had to wait to record it last
>> night. As before the left half of the image is the original and
>> the right half is filtered to an effective 1/4 resolution of only
>> 960x544, though no actual resizing was done.
>> (...)
>> See: http://www.trbarry.com/Leno_1080p_compare_qtr_rez.jpg
Kon Wilms wrote:
> Look at the flag on Leno's chest (yes, that's a flag). Look at the
> coffee cup on his table, vs the one on the left side of the photo. Look
> at the jaggies on the building set behind his head. Those are some bad
> artifacts.
I've viewed the image with ACDSee, which can zoom in by using bicubic
interpolation instead of pixel repetition. The right half of the
picture is suspiciously blocky, which is not something that we would
expect from only 2D lowpass filtering...
I have to say it again: that Tom's method of applying 8x8 DCT and
then setting the 48 higher coefficients to zero is NOT a good method
for simulating down-sampling and up-sampling. It is only a good
method for creating blocking artefacts. Sorry, Tom.
We are trying to simulate the what-if case: first down-sample
1920x1080(p) to 960x540(p), then apply MPEG-2 compression and de-
compression (with 120x70 DCT blocks of 8x8 each, right ?) and then
up-sampling back to 1920x1080(p). I say (p), because this may have
been an image from a 1080i transmission, but with so little motion
the de-interlacing to 1080p is trivial.
There is a big difference between letting high frequencies disappear
through proper down-sampling (with anti-aliasing filtering and all)
or by first cutting the image into blocks and then discarding the
higher DCT coefficients. That is because the block edges themselves
contain higher harmonics, which are not part of the original image.
Discarding these higher harmonics is what causes the blockiness...
Proper down-sampling works over the entire image, not over blocks,
it is therefore immune to blockiness.
Therefore I ask Tom to provide me with the original image, lightly
re-compressed please (that is, a JPEG that is no worse than the
original MPEG), and then I can apply actual down- and up-sampling
to the entire picture and send it back to Tom.
I find it more illustrative to toggle between two pictures (ACDSee
can do that, or we even have a window-wiper program for that), than
to compare two sides of a single image, with a different content.
So, Tom, if you would please send me a single frame ?
Or else Hans, from the EBU, can give me some other HD frame.
Greetings,
-- Jeroen
+-------------------------------+------------------------------------------+
| From: Jeroen H. Stessen | E-mail: Jeroen.Stessen@xxxxxxxxxxx |
| Building: SFJ-5.22 Eindhoven | Deptmt.: Philips Applied Technologies |
| Phone: ++31.40.2732739 | Visiting & mail address: Glaslaan 2 |
| Mobile: ++31.6.44680021 | NL 5616 LW Eindhoven, the Netherlands |
| Pager: ++31.6.65133818 | Website: http://www.apptech.philips.com/
|
+-------------------------------+------------------------------------------+
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at
FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word
unsubscribe in the subject line.
Other related posts:
