LaserDisc Database
https://forum.lddb.com/

3D comb filtering really better?
https://forum.lddb.com/viewtopic.php?f=30&t=4256		 Page 1 of 1
Author:  Guest [ 28 Jun 2014, 03:07 ]
Post subject:  3D comb filtering really better?
Came across some evidence suggesting 3D comb filters can actually remove details in fast motion scenes (action movies e.g.). Has anyone experienced this (or can prove it better yet)?

Always thought 3D comb filtering was better than 2D. Also heard 3d comb filters switch to 2D mode automatically when there is fast motion (automatic stuff kinda irks me). If 3D comb filters only work for static images and cause problems with motion, then I don't see the point of using them on laserdisc.

Here's a link from Compro Tech. themselves suggesting 3D comb filtering can blur fast action scenes on one of their capture cards (one that seems highly recommended by most capturing composite sources). http://www.comprousa.com/russian/product/vmtvultra-3dyc.html

Can sort of confirm this myself. Using a line doubler with 3D y/c separation and it sort of does look like a temporal filter of some sort in high action scenes (without DNR). It still has more resolution to my eyes, vs my 2D 5-line comb filter capture card IMO.
Author:  signofzeta [ 28 Jun 2014, 05:21 ]
Post subject:  Re: 3D comb filtering really better?
The thing with LD is....nothing is always better. Different discs have different optimal settings on different setups. You just have to figure out what you want and go with that.
Author:  substance [ 28 Jun 2014, 05:39 ]
Post subject:  Re: 3D comb filtering really better?  Topic is solved
Adaptive 3d comb filter combs static areas of the image and step back to 2d on areas with motion.

There are 4 registers to deal with. Most devices will not expose these registers to end user for fine tweaking. Most comb filters were designed ntsc broadcast and vhs in mind so they are not always optimized for ld. Those registers are;

Luma coring: threshold for number if pixels the comb filter should consider noise/no motion. Anything under this setting is considered noise even if not static. Anything above is 3d applied if motion detected.

Luma gain: how fast it should switch between 2d to 3d or 3d to 2d.

Chroma coring: same as luma coring but this time only color information is used.

Chroma gain: same as luma but for color.

Analog to digital converter converts video to digital. Each pixel is given a value. Comb filter compares each value on 2 or 3 frames. Areas where value differ is considered motion. Registers above are used to tell comb filter where is motion and where is noise. Noise is bunch of small pixels moving and confuses comb filters. Ld is a very noisy format.

If you set the threshold correct and adjust the filter for fast switching then the artifacts by 3d effect is less appeared. Tge benefits of 3d comb outweighs its issues on ld. The image is much sharper and colors are way better pronounced.
Author:  gumbyandpals [ 02 Jul 2014, 23:11 ]
Post subject:  Re: 3D comb filtering really better?
That's really good info substance. I have a Sony CRT that lets me adjust all four of those parameters you were talking about (via service menu). I only had a vague understanding based on the service manual for what they were actually doing.

I have generally preferred good 2-D comb filters because you have less artifacting from the 2D-3D switch and no motion smear. Cross-color rainbowing doesn't really bother me. However, a properly set 3-D comb filter does produce great results.

The often cited snell and wilcox moving test pattern should not be the ultimate test of a comb filter. Real-world material is a much better test in my opinion. Watch something you know very well and look for the differences between different comb filters. Choose what looks best to you.
Author:  seroxx [ 03 Jul 2014, 20:17 ]
Post subject:  Re: 3D comb filtering really better?
substance wrote:
Luma coring: threshold for number if pixels the comb filter should consider noise/no motion. Anything under this setting is considered noise even if not static. Anything above is 3d applied if motion detected.

Did you mean "threshold for number if the comb filter should consider pixels as noise/no motion." ?

What do you mean by "number"?

Otherwise, thank you for that explanation. Do you have CII? What are your settings there in Comb Filter menue?

Thank you, s.
Author:  substance [ 03 Jul 2014, 21:40 ]
Post subject:  Re: 3D comb filtering really better?
I meant to say number of pixels. The settings on the c2 does not reflect actual number of pixels. Its 0 to 100 in 1 increments.

C2 settings are

8
8
32
45 (or 40,I have to check)
Author:  seroxx [ 04 Jul 2014, 10:28 ]
Post subject:  Re: 3D comb filtering really better?
Both gaines at 8 ?! Why so slow switch between 2D and 3D ? Mine are:
Both gaines at 100
Both corings at 50
What I notice and like is that static parts are very sharp and moving parts are not. Is that the idea of 3D Comb Filter?
Another problem is that I don't have a Crystalio ii remote so I can't adjust the picture from far away. So on very close range moving only slowly. Steps of "1" causes no noticing a real difference. Could you please explain in sort of simplistic way for everybody interested what are 2 following differences:

gain 0 opposed to gain 100
coring 0 opposed to coring 100

and - what is the ideal treshold of c2 for LD picture and, in your opinion, why?
Sorry - I can tweak for hours but I'm very interested in an opinion of a very experienced person.

Another thing - I find VXP Gennom much better than Faroudja which looks very soft and with washed colors. I tweaked with Faruodja but didn't get the proper result. That I find weird because it should be ideal for SD as LD is. Well, it's not. VXP does a really best job de-interlacing, much much better then new Q-deo, and scaling much better then Faroudja. VXP LD-picture is crispy; and then trough Q-deo's fantastic YNR beautifully clean and still detailed (in probably maximum LD - limit). I am really happy with Crystalio II and finally compared it with my Marvell Q-deo. The idea of Q-deo was to make a VP for every possible videoformat ever. It does a great job with HD though, also maybe you tube streams and such. Q-deo for LD is no-go for de-interlacing: there are artefacts. 1080p/24 gives the better detail of all other resolutions but is completely unstable (a problem with recognising LD-signal for clean processing) and not even comparable with detail and stability of VXP Gennom. I would have like to have gotten this analisys before from someone else, but now I give it here for everybody else myself.
Author:  substance [ 04 Jul 2014, 14:32 ]
Post subject:  Re: 3D comb filtering really better?
First forget about 1080p24 on ld. It wont work correctly. You will often have judder whenever there is bad edit and deinterlacer can not lock to all cadences. Stick with 1080p60/50.

Adv chip on c2 will digitize the feed and convert each pixel in mathematical numbers. 3d filter compares mathematical numbers on 3 concurrent frames. If any number differs that means there is motion.

Coring is minimum value where the comb filter should look for motion. If coring was 0, any moving pixel would be motion. Your comb filter would be in 2d mode at all times. If you set your coring too high then even small moving objects will be ignored and applied 3d. The idea is to tell the comb filter the size film grain/noise which act as motion if coring is set low.

If your gain is too high, it will always be in 2d mode. Gain 100 means always 2d. Gain 0 is always 3d. If you set coring high like 40 then gain should be small. If your coring is low then you can set a higher gain.

Start off with default values then adjust coring first. Then play with gain. Disable chroma trap filter. Use dot crawl filter at dynamic.
Author:  seroxx [ 04 Jul 2014, 15:23 ]
Post subject:  Re: 3D comb filtering really better?
Thank you very much! A great help it is, sir! By the way, I'll experiment more with 1080p 24/50/60 on both Crystalio II and Pio AVR. The worth on Pio AVR is: the picture is 20% more detailed in 24 then 50/60. On Crystalio II though the difference is not as much if any, but I'll check it out. You are right about cadences in 24 , picture stutters sometimes. But - since the signal goes trough Crystalio II it stutters very rarely!
Author:  hauntmedoitagain [ 06 Jul 2014, 06:06 ]
Post subject:  Re: 3D comb filtering really better?
To my knowledge, a 3D comb filter will retain the full resolution even in adaptive mode, something 2D filters do not do which is a major advantage.

Also, a good 3D filter will not smear unless it has DNR available to it; my DMR-ES10 doesn't have any artifacts until NR is switched on.
Author:  seroxx [ 06 Sep 2014, 10:42 ]
Post subject:  Re: 3D comb filtering really better?
I have to correct myself about Faroudja. In Crystalio's Faroudja Tru Life menue there are extremely important tweaking options to make the picture great. Better then VXP Gennom for LD.
Author:  substance [ 06 Sep 2014, 15:40 ]
Post subject:  Re: 3D comb filtering really better?
Faroudja truelife is an early form of edge/detail enhancement similar to darbe darblet but much more primitive. It can add artificial sharpness. If adjusted very carefully ld might benefit but have stay in very conservative settings in order to minimize ringing and other artifacts.

There are two issues with Faroudja chip implementation in crystalio 2.

1- its gray scale is shifted. If you look carefully in dark scenes there is a green tint. It is a bug in fw that pms never fixed/never will. You can do a 21 point cms calibration for faroudja setting which will correct the shift to remedy this problem. You need a spectrometer/colormeter and calibration sw for this.

2-it fails more real life deinterlacing tests than vxp. Dcdi is only for video and it works real well but film mode cant lock to every cadence. Vxp is very good in real life and cadence tests, maybe one of the best except anime cadences. Its the short coming of the early gennum chip it can not lock to anime cadences like 2:2, 4:4 etc.. revised gennum chip used in lumagen radiance units dont have this issue. Faroudja fails anime cadences too. If you watch a lot of anime you want a dvdo unit with abt or marvel qdeo. Hqv chips cant lock to anime either.

All pros and cons given, vxp chip is better implemented and should be the choice. Vxp implementation in lumagen is a bit better with lumagens proprietary scaling algorithm. But lumagen radiance units have poor analog section and poor comb filter except their newest radiance 2122 which is $3000 and no less.
Author:  seroxx [ 06 Sep 2014, 20:53 ]
Post subject:  Re: 3D comb filtering really better?
Very interesting; thank you for sharing such detailed knowledge. The best thing would be if you could see this for yourself. I noticed about small cadences problems and the green tint in Faroudja Truelife Film Mode, but the signalflow Faroudja - Q-deo - Reon is making really fine results in everything. The most helpful gadget is so called Waveform Monitor in Pana PT-AT 5000 Beamer (Reon processing) with pressing a default option - it balances the picture to it's correct sharpness and plastic...then I tweak it for the black levels and - voila! I'm speechless. But - I will tweak with gennom VXP again - maybe I can reach something even better...
Author:  substance [ 06 Sep 2014, 23:23 ]
Post subject:  Re: 3D comb filtering really better?
Its not what looks good for me, its whats correct. Enhancement features on displays often play with the gamma curve to boost contrast.

I calibrate my components and displays to rec 708/isf standards with using a spectrometer and a colormeter. All settings on my components are adjusted to output whats on the disc faithfully. Might not be as eye catching to everyone but I am a purist I see what the director intended me to see and how he intended me to see it.
Author:  ertoili [ 08 Oct 2014, 21:19 ]
Post subject:  Re: 3D comb filtering really better?
I think with a good CRT you dont even need a s-video out player
Page 1 of 1 All times are UTC [ DST ]
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/