View Single Post
Old 2nd June 2006, 10:31   #18  |  Link
redfordxx
Registered User
 
Join Date: Jan 2005
Location: Praha (not that one in Texas)
Posts: 863
Quote:
Originally Posted by 708145
I suggest is a filter (DQI.dll)
And what is the difference?
Quote:
I have to think more about that idea. At first it sounds great but at second thought I have some doubts.
I fear it will smooth noisy flat regions too much... any test results regarding this?
Well the question is whether smoothing noisy areas is bad or good;-) But definitely, fine detail was problem in SmoothDeblock2. SmoothDeblock3 is little better... But it does not remove the cause of the problem it only rescues something of what was damaged
Quote:
To sum it up, my plan for a realtime version of this is:
I always thought that coding this script would dramatically increase speed, because is the script many operations are useless just because I have to apply the operation on whole clip. But the idea still needs to be improved.
Quote:
o just use 4 shifts
benefits of 16:
little sharper and less artifact (artifacts appear when are highly quantized hi freq areas... compare SmoothDeblockLayer(quant=6,density=2 or 4))
you can choose any weighting mask you want
Quote:
And here is the acronym
CAPS - Content adaptive detail preserving smoother
Hmmm shouldn't be the word deblock in the name? (reason - search)



I have one more approach to try to implement: First adapting and then merging shifts.
For instance split the frame to areas 32x32 and apply appropriate quant only to that area. It would be milions of crops in script...;-)
For this is needed very good and smooth adaptation mask prepared.

Last edited by redfordxx; 2nd June 2006 at 10:34.
redfordxx is offline   Reply With Quote