Blockiness in a running water scene

[Umen Pich]

Hello Guys,

I wonder if you can offer a trick against this nasty blockiness which I cannot get rid of in a running water scene (creek full with salmon fish)
Apparently the water pattern changes so quickly that motion search algorithms are in trouble.
My original footage is 848x480 pix, MJPEG coded.
I tried h.264 quality profile, 3000kpbs, 3 passes -- bad result. Nasty mosaic of blurred little spots. (this was done by Mencoder264 GUI + Mencoder)
Then I tryed divX, Insane profile 4000kbps , 2 passes - really bad blockiness.

Then I tried Mencoder264 again with a quality preset of 9 (out of 50). This ended up with a filesize bigger then the source and a videobitrate of 26,000kbps, which on top of that couldn't play smoothly (I guess the latter should have been expected)

Back to the beginning -- is this a hopeless case or there might be any tricks I don't know about ?

[jagabo]

Use Divx 1-pass quality based encoding. Set the quantizer at 2 or 3. Or h.264 with CRF or CQP encoding at 18 to 22.

[poisondeathray]

How poor is your original source quality? Can you post a small sample to a free hosting site? (e.g mediafire)

[jagabo]

Be aware though that pure quality based encoding can lead to bitrate spikes too high for some players. The Divx encoder does pay attention to the Certification Profile and will limit the bitrate based on the profiles max. Xvid doesn't do this and will use whatever bitrate is necessary to achieve the requested quality. A really noisy D1 video (like between channel static) can peak over 80,000 kbps. Even x264 at CRF 22 will generate over 50,000 kbps with this type of video.

[Umen Pich]

Thanks for the answers !
The original quality is good enough.
I will definitely post a piece of it, when I go back home (I am at work now)

[filmboss80]

Also look for a profile for motion estimation in your divx encoder. You'll want the highest quality setting.

You have identified the one biggest weakness in the divx/xvid compression scheme: it does not handle images of running water very well.

[Umen Pich]

Here is a short clip:
http://rapidshare.com/files/164889079/salmon.MOV

I admit I saw some blockiness in the original while looking at it frame by frame. But if you look at the motion picture it is much better and smooth than the one which you get by 4000kbps divx

[jagabo]

Yes a shot like that will take a lot of bitrate. As a test I encoded it as Xvid with a target quantizer of 3. With most video that is a good balance of file size and image quality with only a little macroblocking. The result looks fine but GSpot reports the bitrate as 18,000 kbps. With x264 at CRF 22 it came out around 13,000 kbps.

[poisondeathray]

Yes you will see pixellation in many frames in the source; that is the main problem. IMO, this is more of a case of pre-processing than encoder settings. I suggest you do some processing with avisynth.

There are many ways to tackle this. I prefer sharper images, so I denoised & sharpened after, perhaps others will suggest better solutions

QTInput("salmon.MOV")
ConvertToYV12()
Deblock(quant=40)
Undot()
LimitedSharpenFaster()

original


deblocked


deblock undot lsf



Here is the clip ~6.6MB (~6500kbps video)
http://www.megaupload.com/?d=CBC7GLW5


As for encoder settings, you could raise the inloop deblocking in x264, but that alone will have less of an effect than the pre-processing in this case. I did turn off psy-trellis and psy-rdo, set deblocking to 0,0, and used crf22.

[jagabo]

this is more of a case of pre-processing than encoder settings.

Nice! Eliminating all that high frequency noise in the source cut the bitrate requirement substantially.

[Umen Pich]

Very good idea indeed. Thank you !