Should I temporal denoise before or after IVTC?

[Trixter]

I have some film->NTSC content recorded off of cable TV that I'd like to clean up. What is the current conventional wisdom regarding IVTC and noise removal: Which should be done first? I can't decide on my own which is better.

For example, would performing IVTC first help or hurt the noise removal, since there is less temporal information to denoise across? Or is IVTC best left until the signal has been cleaned up -- but would the cleanup screw with the IVTC pattern detection? Help?

(If you're curious, there's no rainbowing or dot crawl; just analog broadcast noise, and a little film grain present in the original. I'm using telecide/decimate for IVTC, and degrainmedian for one noise removal pass with fft3dfilter doing a light luma-only 2nd pass.)

[Mug Funky]

hmm. looks like the seesaw principle :)

i'd be inclined to IVTC first. it really depends on how high the noise floor is, and how even it is.

use your usual IVTC settings, and if you're getting more false matches than usual (look around mouths in static scenes), then try a gentle denoiser like degrainmedian and see if that helps or hurts. my gut's telling me that it'll hurt it, but you never know.

[Trixter]

i'd be inclined to IVTC first. it really depends on how high the noise floor is, and how even it is.


Good point. The noise, since it's 30i broadcast, is fairly even/uniform/predictable. But I'm still going back and forth over whether or not denoising first is the better option...

[JoeShrubbery]

I look at it as a speed/quality tradeoff. IVTC filters and deinterlacers probably work better with cleaner input, but the degree to which that affects things is up in the air and probably dependent entirely on the source capture you're working with.

Personally, most of my encodes from TV caps get denoised last, after either IVTC (or deinterlacing of pure 30i video) AND my resizing. I believe that I'd probably get better results denoising earlier in my scripts, but pushing it down to last pumps up my encoding speeds a few FPS (fewer frames to process after decimate, smaller frame size after resizing) and the results still look good enough to me for my purposes. If there's something I want to take extra care with, denoising gets moved waaaay up to as early in the script as possible, the rationale being that any subsequent filters will probably benefit from having a cleaner input.

Ultimately, all I can really suggest is to try it both ways and then compare. Let your own eyes be the judge of it.

[Chainmax]

IMO you should IVTC first unless the noise is so bad you fear it might confuse the IVTC method. You can always turn up TDecimate's "nt" switch, or whatever equivalent switch the other methods have.

[Trixter]

Well, after dorking around with it for more than 2 hours :-) I found that a combination actually worked best for my footage. I do the simple denoise before the IVTC, then the stronger stuff afterwards. Coupled with some slight AGC, I'm really happy with the results.

Here's the code I'm using:


DegrainMedian(interlaced=true)

Telecide(guide=1,post=2,show=false)
Decimate()

HDRAGC(debug=0, max_gain=2.0, max_sat=2.0)

FFT3DFilter(sigma=4, plane=4, interlaced=false)

...which turns this:

http://img322.imageshack.us/img322/705/ivtconlyfe5.th.png (http://img322.imageshack.us/my.php?image=ivtconlyfe5.png)

...into this:

http://img322.imageshack.us/img322/2669/ivtcdenoisehgragczu7.th.png (http://img322.imageshack.us/my.php?image=ivtcdenoisehgragczu7.png)

The choice of sigma=4 in fft3dfilter is a bit strong, I might back that off; but since the original is 3500kbps MPEG-2, it has the nice side-effect of deblocking the source and also dealing with the occaisional chroma "blotches" created if the AGC gets too strong.