Help with DCT Blocks

[Beek]

Hi,

I am a relative newcomer to encoding using avisnyth scripts, and am having a great problem encoding my digitial tv captures due to what I assume are DCT Blocks (ugly undulating and squirmy areas in near dark areas).

To capture I use the Hauppauge WInTV-PVR-USB2 hardware mpeg encoder at 720x480 at 12000 kps. I am very pleased with the captures, but they exhibit DCT blocks either from the digital cable source or the mpeg2 capture.

I have tried many different noise reduction scripts such as deen, fluxmooth, convolution3d, fft3d but not really happy with the result, as they take out too much detail from the overall picture.


Blockbuster seemed to be what I was looking for, but from my understanding it doesn't remove these dct areas from source it only keeps pristine source from having them due to the encoding process.

If my assumption is correct (that blockbuster wont correct the blocks in the source), could anyone suggest an appropriate filters and its usage to deal with the problem I am having keeping in mind that I want to preserve as much detail as possible in people faces and other items in the foreground.

If my assumption is incorrect (that blockbuster is perfect for my problem) could someone suggest a script using blockbuster that would be ideal form removing the DCT blocks in my source.


Any help would be much appreciated as I have been trying to deal with this problem for some time now. I am almost at the point where I want to ditch the mpeg encoder and buy one that can caputre in huffy avi.


Cheers, Steve

[dukey]

the blocking is probably already in the source before you encode it

[danpos]

@Beek

Firstly you've to attack the blocks present into video. For this you've to use a deblocker. After that, you can to use Blockbuster in order to encoder doesn't produce DCT blocks into future MPEG2 videostream. For Deblocker you'll can take a look at this this topic (http://forum.doom9.org/showthread.php?t=110936) and others (:search: ).

Cheers,

[foxyshadis]

The direction you should be looking toward is "deblockers", such as Deblock, MDeblock, FunkyDeblock, Deblock_QED, and the promising but unfinished ReDeblock. Additionally, if you're loading from mpeg2source, you can specify adaptive deblocking via cpu=4, or you can use ffdshow deblocking (or whatever your decoding filter is) if loading from directshowsource.

[Dan beat me to it. :p]

[Mug Funky]

i just had an idea for using frfun (at fast and agressive settings) instead of a lowpass to preserve edges more in a deblocker...

might have to hack it into funkydeblock to see what comes of it... frfun has been working very well with seesaw for sharpening, and i like the "blurred with preserved edges" idea, and the speed.

but for now deblock_QED and similar stuff is probably the way to go. i mostly use my own deblock (basically a lowpass-protected version of blindpp) for anime and some natural video. it fails on extreme blocking though, and can kill some low-level grain details.

[redfordxx]

Lowpass filter means that it kills high frequencies=detail?a lowpass-protected version of blindpp
And what does this mean?

[Beek]

Thanks for the replys all.

[Mug Funky]

@ redfordxx:

basically means it does a lowpass (well, not strictly - it's a resize-upsize blur because it's fast and will do for the purpose), subtracts the original to create a highpass, runs blindpp on the highpassed clip and then subtracts again with the lowpass to restore the original clip (plus deblocking).

to be honest i don't know why it works any differently at all to regular blindpp, but it seems to catch more blocks and preserve (slightly) more detail.