Convolution3d with User-Defined Matrix?

[primitive]

It might sound kind of wacky, but I have a clip I think the matrix


0 0 0 1 1 1 1 1 1
0 0 0 1 1 1 1 1 1
0 0 0 1 1 1 1 1 1


would be effective on. What's happening is that the clip is a sequence of credits where every single frame of the scrolling credits has combing artifacts, but playing alongside is a small window with animation in it which is combed using a different scheme. TomsMoComp(0,5,1) takes care of most-all the combing, but I still have edge artifacting (from all the nasty combing) and those artifacts you get from making intraframe comparisons (like I posted a thread about earlier).

If you step frame-by-frame through this material, you see that on those frames that you have heavy intra-frame artifacting, the artifacts correspond with regions of color in the previous frame, and in most cases the next frame is just like the previous frame, except that it's free of artifacts.

I'd try to apply a temporal smoother to this clip, but I thought a con3d with s user-configurable matrix would be more generally useful.

Let me know what you think.

-p

[vlad59]

It's true that I totally forget the use of user-defined matrix...:rolleyes:

But if you simply want to test (= speed is not important for your sample), you can try the SimpleConvolution3D (or SimpleConvolution I'm not sure of the name) filter. It's hidden in the Smooth in Time and Space thread.

If the results are interesting then I'll think about optimizing a little SimpleConvolution3d or maybe if I have a wonderfull idea adding it to Convolution3d.

Let's me know if you don't find it, I'll look in my hard disk.

[primitive]

I found the filter, but sadly the thresholds must be set way too high before any noticable reduction of the noise ocurrs, and when the thresholds are that high the filter starts taking huge bites out of the desired video.

It was worth a shot, though :(

I'd still like to test a full Convolution3d filter with the user-defined matrix, though. I doubt I'd be able to make anything of it, but it's always worth a try.

-p

[vidiot]

Could it be that there is a VD filter for Convolution3D where you can put in a user defined matrix ?
(General Convolution 3D / Gunnar Thalin)
Maybe it´s good enough for testing your settings.
But it has a 3x5 Matrix...

Harald