MC temporal denoise camera before Deshaker?

[fvisagie]

Hi All,

I'm working with camera footage that needs both deshaking and low light noise removal. I've got the deshaking sorted and MC noise removal is well on its way, but which sequence is theoretically best? I know "suck it and see", but sometimes the effects are so subtle or only show up in the footage after your test clip (and on my machine each experiment takes so long!) that I'd really appreciate your guidance up front.

Here's a sample of the camera shake I'm working with:
http://www.mediafire.com/file/qcx3ix71kbh9ovp/Sony_DCR-TRV330_PAL_DV_Sample.avi

Here's a sample of low-light footage:
http://www.mediafire.com/file/3yq98m86d2e8q5o/Sony_DCR-TRV330_PAL_DV_Low_light.avi

In my simplistic way I'm thinking that the better sequence might be Deshaker > MC noise removal. Reason being that Deshaker breaks up the original motion path of the footage, therefore making the motion-induced noise more random and easier to isolate? I'd very much appreciate your inputs on this issue.

Edit: Just to point out, in this case Deshaker does not need the footage cleaned beforehand. It does a great job determining and compensating motion with the footage as it is. So it's purely a question of when is MC temporal denoising expected to work best - before Deshaker or after?

As always, many thanks!
Francois

[sanlyn]

The camera shake video doesn't need MCTemporalDenoise (overkill). Not enough noise to worry about. It's interlaced, so will look smoother playing on the computer if you view it in an app with decent deinterlacing (such as VLC Player with deinterlace turned on, or VirtualDub).

The second video will need more than MCTemporalDenoise if you want to retain any detail. It's mostly noise.

[fvisagie]

For this particular project's footage QTGMC() > Deshaker > TemporalDegrain() worked beautifully. Thanks for all the help and suggestions!

Edit: Agreed, that particular camera shake clip doesn't need denoising. I normally view interlaced content in one of VirtualDub's deinterlacing display modes as you suggest, but other parts of the footage were really grainy, even after QTGMC() for Deshaker.