Log in

View Full Version : How to get de-blocking to work at low luminance


gooberon
10th April 2008, 11:35
Does anyone notice the blocking a low luminance levels when using x264? Is there any way to get the deblocking to work at low IRE levels? Or is this not what it was meant to do?

Admittedly its due to the noise in the source but sometimes its there no matter how good the source is due to film grain or bad gamma setup in the transfer.

Any clues anyone?

Dust Signs
10th April 2008, 18:22
Hello,

this is due to the fact that the human visual system is more sensitive to (quantization) noise in dark areas of a picture, so it's not x264 specific and the filters should just work as they are supposed to.

Dust Signs

gooberon
14th April 2008, 09:06
I guess theres always going to be a tradeoff somewhere. Is there a detailed explanation, in one place, that defines each of the parameters in x264.exe.

Dark Shikari
14th April 2008, 09:34
Things you can try:

1) Increasing the --deblock parameter to 1:1, 2:2, or higher.
2) Raising the AQ strength parameter.
3) Using gradfun2db on playback, or before encoding.

gooberon
14th April 2008, 09:58
DS,
gradfun2db (in ffdshow) works very nicely on decodes, now if only I could get ffdshow onto the PS3 ;-P.
I'll give the AQ a try.

CruNcher
14th April 2008, 14:27
http://forum.doom9.org/showthread.php?p=1114782#post1114782 <-- i guess you mean this visual transcoding problem?

@gooberon
Using Darks AQ and Calibrating the viewing Device to the source works wonder for me it completely eliminated this problem on the PC for lossy content :) (thx also to LG's wonderfull F-Engine (manually calibrated not using the Film Preset) (Post Processing in the LCD))

Mr VacBob
14th April 2008, 22:16
Video Y isn't really perceptually linear - it's pretty close along most of it, but changes in darker areas can be much more visible. You can add dithering and use VAQ, but without dither +/-1 differences can be visible.

gooberon
25th April 2008, 02:32
The AQ works quite well but it tends to wash out detail. A side effect I assume of the bit redistribution. I maxed out the AQ so I might pull it back a touch.

Indeed this http://forum.doom9.org/showthread.ph...82#post1114782 is the problem. Post processing may be the answer but I hate that, its like adding MSG :-p