Interesting thread about DeNoising

[bb]

Hi DV lovers,

just wanted to point you to this thread, as noise filtering is very important for you:

http://forum.doom9.org/showthread.php?s=&threadid=29829&perpage=20&pagenumber=1

Have fun!

bb

[theReal]

I feel free to add this thread: http://forum.doom9.org/showthread.php?s=&threadid=29359

Acaila's script for seperating chroma and luma for filtering is worth a try (as well as Marc FD's cnr2-filter).
Doing a light temporal smoothing on luma (or no filtering at all), and adding a stronger filter for the chroma (a spatial smoother, if you have enough cpu power and time) really does produce much less visual artefacting than filtering everything together, but it still works very well for making the video compressible.

Adding a strong smoother on the chroma can do wonders for homevideos of birthdays and similar occasions (not enough light, much chroma noise especially on white walls...). However it does not blur the video because the luma remains unchanged.

[vivaatloni]

Thx for the links.

I am interested in the convolution3d setting. What settings have you (BB) used with DV encoding.

I am gonna test some tonight. I test the high quality with lot of noise.

[Bandung]

bb,
Thanks for the thread.
I had started to read this thread when it first began, and somewhere around page 12 or 13, I stopped. It was great to pick it up again and to be reminded how convoluted 3d in its present format, got its start.

Thanks again.

[pyropir]

Unfortunately the thread is not only extremely long but it also contains very little information about c3d's usefulness for DV. From what I've read, it seems really quite important to get the settings right, but I have no idea what settings to use.

I've used the Anime Low Quality (1, 8, 16, 8, 8, 2.8, 0) and Anime Bad Quality (1, 12, 22, 8, 8, 2.8, 0) settings, but I'm very doubtful about wether these are good settings. Unfortunately I didn't have the time to do proper tests. I understand, of course, that each film is different, and that there is no way around testing.

Can anyone suggest settings they have used with satisfaction, or at least give a good idea where to start optimising settings?

pir

[bira]

I am using the following to process my canopus DV avi.



AviSource("file.avi")

FixBrokenChromaUpsampling

separatefields()
vid_e=selecteven()
vid_o=selectodd()
vid_e=convolution3d(vid_e,0,4,4,4,4,2.8,0)
vid_o=convolution3d(vid_o,0,4,4,4,4,2.8,0)
interleave(vid_e,vid_o)

weave()

ResampleAudio(44100)

[vivaatloni]

Thx for the info BB and pyropir.
I have succesfully used the setting for bad dvd source and this output file is better than when I don't use convolution3d.

I really don't understand your settings BB. Why doe you deinterlace the movie?
Can you explain a bit more what these settings are doing and what your output format is.

[bb]

@vivaatloni:
I guess you refer to the PM you sent to me, where I recommended Convolution3D(0, 4, 8, 4, 4, 3, 0) or Convolution3D(1, 4, 8, 4, 4, 3, 0). The first one leaves more noise in the video, whereas the second one blurs a little more.

bira's script is fine for those who want to keep the video interlaced. I recommend to do this for high bitrate DVD target formats; for SVCD you better deinterlace because of the better compression and an improvement of the overall quality. For DivX you must deinterlace, because the codec doesn't support interlaced video. XviD does, though, but I have no experience with that.

My preferred output format is DivX @ q3, Ogg Vorbis, OGM file format. I'll probably switch to DVD as soon as I get a burner. Note that I do not need to deinterlace, because my camcorder supports progressive scan.

I'm a little curious how interlaced and progressive modes compete for on high bitrate DVDs.

bb

[bira]

Originally posted by bb
Note that I do not need to deinterlace, because my camcorder supports progressive scan.



You're lucky. :)

[vivaatloni]

Thx for explaining BB. But de-interlacing makes takes a lot of quality in my testing (smart de-interlace).

I think I am gonna try the method of Bira. If he can explain wat it means. So Bira why don't you use 1 convolution setting but some sort of script and can you explain every step to me?

[bira]

I'll try but I am no expert.

The names are pretty much self explanatory.

FixBrokenChromaUpsampling - it will get rid of the canopus decoder bug as described in this thread (http://forum.doom9.org/showthread.php?s=&threadid=33526)

Separatefields - since I am not as lucky as bb, my dv avi is interlaced so I need to separate each field of a frame in order to apply convolution3d.

Selecteven and selectodd - creates a clip with only the even fields and another with the odd fields.

convolution3d - applies convolution3d to each clip

interleave - interleaves frames from the two clips. even - odd - even - odd and so on.

weave - interlaces back the avi.

resampleaudio(44100) - to fool CCE's bug


If I said any non-sense please shout!

Hope it helps.

[vivaatloni]

Thx for explaining. I am gonna give it a try. I am using dvd2svcd. Does the following script gonna work or do I have to load plugins like seperatfields, selecteven etc.

My script looked like this one:
LoadPlugin("C:\PROGRA~1\DVD2SVCD\MPEG2Dec\mpeg2dec.dll")
LoadPlugin("C:\Convolution3D.dll")
AVISource("D:\DVCAPT~1\noor0090.avi")
ConvertToYUY2()
Convolution3D (0, 6, 10, 6, 8, 2.8, 0)
BicubicResize(480,576,0.00,0.60)

Can I just change it to this one or do I have to load new plugins. And where can I download them?
LoadPlugin("C:\PROGRA~1\DVD2SVCD\MPEG2Dec\mpeg2dec.dll")
LoadPlugin("C:\Convolution3D.dll")
AVISource("D:\DVCAPT~1\noor0090.avi")
ConvertToYUY2()
separatefields()
vid_e=selecteven()
vid_o=selectodd()
vid_e=convolution3d(vid_e,0,4,4,4,4,2.8,0)
vid_o=convolution3d(vid_o,0,4,4,4,4,2.8,0)
interleave(vid_e,vid_o)
weave()
BicubicResize(480,576,0.00,0.60)

Last question. Do I need the FixBrokenChromaUpsampling cause I am using the mainconcept dv codec?

Hope you can help me. Thx in advance!

[bira]

Originally posted by vivaatloni

Last question. Do I need the FixBrokenChromaUpsampling cause I am using the mainconcept dv codec?


The answer is no, but you should read the thread I mentioned.

Why are you loading mpeg2dec.dll? You don't need any other plugins besides Convolution3D.dll.

[bb]

What bira said is all correct. vivaatloni, if you succeed in producing an interlaced SVCD having a better quality than going the progressive way, please let me know. But I really doubt it: you should get more macroblocks, and more mosquito noise as well, because your bitrate is limited by the SVCD standard.

bb

[vivaatloni]

Thx bira en bb. I am gonna try this setting tonight.
The reason why I don't use de-interlace is because the threads I read about them. All say it makes the image less sharp and that is somthing I realy don't want.

I also was a big fan of cce, but in some test I performes a lot less than tmpg. So I switched to tmpg instead of cce I earlier used (only for dv material though). What deinterlacing option are you using BB?

dvd2svcd loads the option mpeg2dec.dll automaticly. It also performs ConvertToYUY2() and I don't know why he does that either.

You gonna hear from me about the results.

[bb]

Originally posted by vivaatloni
The reason why I don't use de-interlace is because the threads I read about them. All say it makes the image less sharp and that is somthing I realy don't want.
Even if I repeat myself: Your resizing method will probably blur your video more than deinterlacing.

What deinterlacing option are you using BB?
I'm switching my camcorder to progressive scan :D
No, seriously, I use AviSynth with Decomb, Fielddeinterlace(), just the defaults - they work fine.

dvd2svcd loads the option mpeg2dec.dll automaticly. It also performs ConvertToYUY2() and I don't know why he does that either.
You definitely don't need mpeg2dec.dll, because you don't decode VOBs from a D2V project. You don't need ConvertToYUY2 either, because DV's colorspace is YUY2 already (but it does no harm).

bb

[vivaatloni]

@BB
Ok. thx. What resizing method do I have to use to less blur the image?

@Bira
I also tried the canopus codec and it does produce less noise then the mainconcept. Thx Bira, from now on this is the codec I am gonna use.

[OvERaCiD23]

Originally posted by vivaatloni
@BB
Ok. thx. What resizing method do I have to use to less blur the image?


LanczosResize is the sharpest resize (I've heard of) w/o any ringing effect (or at least much less than BicubicResize @ 0,0.75). Lanczos is built into AviSynth 2.06, or you can download it separate.

[bb]

Or use a sharp bicubic resize, which will process faster than Lanczos: BicubicResize(480,576,0,0.75).

bb

[OvERaCiD23]

I'm not sure if I'm the only one, but I can care less about encoding time versus quality. Why encode if not for high quality? You'll have the quality forever, so what's another hour or so of encoding time? Unless of course, it reduces processing to less than 4fps, then it's a factor. But the difference between 12fps and 14fps with Lanczos vs. Bicubic (respectively) has no effect on me. The only things I really consider are compressibility and image quality for video. Slow computers deserve punishment for encoding slowly =) (I'm using a laptop w/ a P3 900mhz, so I'm used to fairly slow encoding).

[bb]

I don't care much about encoding time either, but if you need another hour without seeing a difference, then I prefer faster encoding.

There's no visible difference between sharp bicubic and Lanczos resize, except maybe if you view the video frame by frame at 4x magnification. But then there's still much more impact on video quality by the chosen encoder and their settings, or by the chosen filter chain. The DV compression itself introduces more artifacts than bicubic resizing (in fact downsizing can reduce some amount of these artifacts). I do believe Lanczos resizing is valuable for high quality still image resizing, but for video it's simply overkill.

Just my opinion.

bb

[theReal]

I find that Lanczos Resize reproduces colors better than BicubicResize. However I only use it after convolution3d and if the compressibility is good enough anyways. For DV I guess I will never use it because DV video needs every bit of compressibility you can get...