On Fri, 13 May 2005, Dik Takken wrote:
> out, using yuvdenoise can get you pretty close to that goal. It seems that
> you can treat the luma data quite agressively without degrading the
> appearance of the image.
Hmmm, ordinarily treating the luma aggressively blurs/softens the
image and loses detail quite quickly. Chroma on the other hand
can be (and often is ;)) filtered heavily because the eye is less
sensitive to small changes in color.
> > Did you use the normal/default mode of yuvmedianfilter or the '-f'
> > averaging mode? Something like "-t 0 -f -R 4" might do a reasonable
>
> I didn't try -f, will try that next time.
And with yuvmedianfilter you can, as you no doubt know, selectively
disable luma and chroma processing.
yuvmedianfilter -f -w <weight> -r <luma radius> -R <chroma radius>
Default (center pixel) weight is 8. For minor touchup work I've
found setting the weight to 12 or higher works very well (especially
on animation).
Add a "-t 0" and leave out the "-r" to do the chroma only.
> That is only in low light conditions, and only in the shadows. The groups
Of course. That's where a small CCD shows its limitations most
noticeably - in brighter scenes the problem's masked. Also the
size of the CCD comes into play with the dynamic range and effective
horizontal resolution. The DV format is claimed to be up to 500+
lines of resolution, the better camcorders get around 530, but depending
on several factors (including the size of the CCD) the real resolution
can be much lower (350 to 400).
That's why the (much) more expensive camcorders use 3 1/2 or 2/3"
CCDs and have controls to increase the gain of the sensors :)
> > Ah, but "-t 0 -z 0" doesn't disable the luma processing in y4mdenoise,
> > does it? I think those values cause the most exhaustive time
>
> Ah, yes that would explain it all. Somehow I assumed that yuvdenoise would
> be so wise not to try denoising with an error threshold of zero.
Yep. With a threshold of 0 you've told y4mdenoise to search for
an *exact* match - no error threshold so the program has to do more
searching.
That's why y4mdenoise -t 4 -z 3 runs faster than the less aggressive
"-t 2 -z 1"
> > to help reduce the 'sandpaper' or "boiling sand" effect - something
>
> The boiling sand effect is there indeed. I just had not decided which tool
> I wanted to use to reduce that kind of noise, either yuvdenoise or
> y4mdenoise. I guess my choice has been made. :)
yuvmedianfilter will, I think, be more effective than the denoising
filters. Actually a combination of both yuvmedianfilter and
one of the denoising programs will be the best.
The "noise killer" (Final Cut Pro plugin) from CGM:
http://www.cgm-online.com/eiperle/cgm_screenshots_v3_e.html
seems to be mild (in the sense of not losing detail) but effective
at reducing the 'grain/boiling-sand' effect. Doesn't eliminate it
but every little bit helps ;)
Over the past few weeks I've found that using 'yuvmedianfilter' _before_
the denoising filter works best. I do a lot of processing of
animation and the large flat areas of solid color are extremely
difficult to clean up with the denoising filters alone. What is
producing the best results so far is normal yuvmedianfiltering on the
chroma then luma averaging with heavy center pixel weight, then
y4mdenoise:
(smil2yuv ...) | \
yuvmedianfilter -t 0 | \
yuvmedianfilter -T 0 -f -r 1 -w 12 | \
y4mdenoise -t 3 -z 2 | \
...
It's also possible to do the averaging/fastmode on the chroma
(with a different radius and center weight) with "-t 0 -R N -w M".
I think you're in for a lot of experimentation - there are a lot of
options and combinations to try.
Have fun and Good Luck!
Cheers,
Steven Schultz
-------------------------------------------------------
This SF.Net email is sponsored by Oracle Space Sweepstakes
Want to be the first software developer in space?
Enter now for the Oracle Space Sweepstakes!
http://ads.osdn.com/?ad_id=7393&alloc_id=16281&op=click
_______________________________________________
Mjpeg-users mailing list
Mjpeg-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mjpeg-users
