GaussResize in DV upscaling?

[zambelli]

When upscaling SD video to HD, most people tend to focus on sharpening the video in order to "fake" extra image detail. But I find that when upscaling DV video, most of the time you just end up with enhanced artifacts: low-light grain, jaggedy edges, stray pixels, etc.

So here's a somewhat counterintuitive idea: how about, instead of sharpening SD video to make it look like HD, you blur DV video to make it look less like DV?

Here's an example:
LoadPlugin("C:\Program Files\AviSynth\filters\TDeint.dll")
LoadPlugin("C:\Program Files\AviSynth\filters\EEDi2.dll")
LoadPlugin("C:\Program Files\AviSynth\filters\AddGrainC.dll")

AVISource("DV.avi", pixel_type="YV12")
AssumeBFF()
interp = SeparateFields().EEDI2(field=-2)
TDeint(mode=1, edeint=interp)
GaussResize(1280,720, p=20)
AddGrain(15,0,0)

The goal here is to hide the DV artifacts at the expense of detail.

AddGrain() is used to further hide the artifacts and create an illusion of detail.

The effect, in my opinion, is more "filmlike" and more reminiscent of 8mm than DV.

And here's my question: is there a better way to achieve this goal than using GaussResize()? Is there a way to hide DV artifacts without erroding as much detail?

[Didée]

To my understanding, you are asking whether using a denoiser is better instead of using a gaussian blur?

[IanB]

I guess what you are advocating is to low pass filter the upsized image at DV resolution upper spacial frequencies. i.e approx 540x368 ntsc or 540x442 pal

[R3Z]

I would have thought motion compensated denoising with addgrain tacked on the end would (well is in my eyes) be better than gauss resizing.

I find that a gauss blurred and then sharpened addgrain mask of about (5,0,0) gives a film like effect as well.

Thats my opinion anyway :)

[Chainmax]

It seems to me you're asking two questions in one:

If you want to eliminate artifacting, it would stand to reason that using dedicated tools (like RemoveGrain or DeGrainMedian at low settings for stray pixels, EDI resizing + downsizing for aliasing, MVDeGrain2 for noise, etc) would be better than using a general purpose blur.
If you want to create a film effect, there's an enormous thread regarding that: IIRC, it involved deintelacing, using TemporalSoften and grain adding.

If you want both combine the two methods.

[scharfis_brain]

at first, you should deinterlace using tdeint with tmm and nnedi.

[zambelli]

Denoising can certainly help eliminate chroma noise and CCD noise, but I find that even the best denoising can't make DV video not look like DV - if that makes sense.

Scharfis: I've only casually been following the TMM discussions. Is Tdeint+TMM+NNEDI now considered to be superior to MCBob?

[Terranigma]

Is Tdeint+TMM+NNEDI now considered to be superior to MCBob?

No. Faster? Yes. Mcbob is still the better deinterlacer/bobber. The only thing that yadifmod and tmm does a better job at than mcbob, is detecting static areas. All avisynth deinterlacers/bobbers, in some cases causes artifacting, whereas mcbob causes none (at least for me anyways).


---
btw, if you're going to compare tmm+nnedi against mcbob, then the comparisons must be made with mcbob that also uses nnedi, which you can grab from here (http://forum.doom9.org/showthread.php?p=1052018#post1052018). :D