Spaced Ranger said:
@ AntcuFaalb
Some time back, you were looking into 2D/3D comb filtering. What is the status of the "phase-inversion trick" and a capture card that properly implements both 2D and 3D comb filtering without AGC problems? Are you using both these in the THX 1138 laserdisc captures? (If so, would you post some pictures & descriptions of the work's application & improved captures?)
BTW, how goes the development of the 3D motion-adaptive comb filter?
1. The "phase-inversion trick" turned out to be a dud. After learning A LOT more I realized that you need to flip the phase of the luma, but the phase of the luma is already set because LD is a composite signal. It can hypothetically work if you pass the signal through a comb filter and have it flip the luma phase 180º between captures. Unfortunately, this would still rely heavily on the quality of the comb filter used and the "phase-inversion trick" in question would just create rainbow-sized and dotcrawl-sized holes in the separated signal that may or may not be pleasing to the eye. It's probably best to avoid this altogether.
2. My quest to find the perfect capture card stalled, unfortunately. Every one I looked at had at least one of the following problems: a crappy comb filter, "AGC Gone Wild!", poor chroma resolution (<1.5MHz), a not-precise-enough clock, etc. I eventually settled on the Blackmagic Intensity Shuttle because of its rock-solid clock, excellent chroma resolution, and lack of other problems. Unfortunately, it sports a crappy comb filter, so I only pass s-video into it and rely on an external comb filter to do the job. In this case I'm just using the internal 2D comb filter in the Runco LJR-II. It's good by modern 2D standards (I've done comparisons) and holds up to typical LD noise rather well. FWIW: Mark S. Brasfield (of MSB—the company that modded the LX900 for Runco) tweaked it himself.
I really can't afford a good external 3D comb filter and the less expensive ones are coupled with deinterlacers (including IVTC) and scalers and it's rarely possible to turn those functions off to just pick up the combed 480i signal. Turning those function off is necessary for me because, for instance, the IVTCer can screw up with an unexpected change in cadence and then... well... what do you do to fix that? Plus, from what I've seen, the scalers have nothing on the NNEDI family due to realtime constraints.
Exception: I'd love to work with a Crystalio II VPS-3300/3800 if anyone is feeling generous. ^_^
3. Slow. My use of optical flow turned out to be a bad idea, so I'm now working on a good pixel motion estimation algorithm and I haven't really designed one that's up to my standards (in terms of accuracy) yet.