Log in

View Full Version : Anime filter recommended for derainbow


MasterYoshidino
25th June 2003, 08:23
Some may notice that certain anime that is transcoded from old film/vhs such as oh my goddess or neon genesis evangelion will notice some funny "rainbowing" on edges. This is due to the cable interference when capturing. This rainbowing causes bitrates for mpeg4 codecs to reach 2000+kpbs when encoding high motion due to the rainbowing, which does nothing but reduce asthetics. The best way to filter this noise (when I mean best I mean nothing compares to it) is SmartSmoothIQ. (IQ NOT HIQ)

It surpasses temporal filtering on Avisynth filters as they rely on static edges to filter rainbows, while IQ will filter even hi motion scenes.
A decent way to encode would be to just run the telecide and mpeg2source filters in avs script and then load it and load SmartSmoothIQ at least twice. Three times is better, anymore will be overkill.
Save the AVI as a huffyuv so that you can apply avisynth filters to the temp IQ filtered file, sans rainbows ;).
Below are screenshots taken from Oh My Goddess OVA intro credits, where any other filter fails due to the hi motion preventing temporal median/average filters from doing any derainbowing.

Unfiltered, IVTC only (http://masteryoshidino.adultbouncer.com/images/goddess_ova_unfiltered.png)
IVTC, followed by 3 passes of 11 radius SmartSmoothIQ (http://masteryoshidino.adultbouncer.com/images/goddess_ova_filtered.png)

Dark-Cracker
25th June 2003, 11:57
hum i think your encode will made very very long time why don't use somethink like :

MergeChroma(AntiBlink(10,debug=false).SpatialSoften(7,0,50).blur(1.5))

i think u could have good results in less time.

++

MasterYoshidino
25th June 2003, 16:29
because it does not do the job as well as IQ can. I already stated that credits are high motion and they can not be filtered by temporal filterings. I have tried mfRainbow, AntiBlink, GuavaComb, even insanely high Deen a3d but none touch the rainbows in the credits.
IQ preserves the MOST detail, it even preserved the combing artifacts that Telecide failed to remove even with post=true !!! (try zooming in my png pictures at least 6x and look to the right of Urd and you will see that both pictures have the combing)
A/W 20 hours to encode a 30 minute temp huffy is nothing. (2.4GHz P4)
I am expecting compression values of constant two from the rainbowed 450MB (for DRF2) to 280-320MB (the rainbowing is what causes insane bitrates)

Kyo
4th July 2003, 06:15
I don't remember well now, but the filter DctFilter.dll from Tbarry (great filters BTW) will filter those high frequencys and reduce the efect of those in the final compresion, but not clean at all, Some time ago(I used the YUV version of the filter) I use this filter(just to test) with a trigun ep, and the final quality was superb, for just 160Megs and 640x480 res. But I remember that the YV12 version don't have the same "strange" efect.

/me Think that we need more tests on this filter.