Log in

View Full Version : Filter idea: Motion adaptive temporal filter?


ronnylov
14th September 2002, 08:31
Hi all!

I wonder if it is possible to have motion detection to modulate the strength of a temporal filter? This could be useful for analogue captured clips with still pictures like cartoons but also on normal movies.

We have the MAM filter which toggles between low motion filtered clip and fast motion filtered clip. We have CopySame and the dup filter that duplicates identical frames where no motion is detected. But is there a filter that averages identical (or almost identical frames) as long as no motion is detected? With a motion treshold it could be set up to do temporal smoothing of all no-motion-frames until motion is detected. I guess this would require some kind of frame buffering of the almost identical frames until motion is detected and then the filter is applied on the "almost identical frames". At high motion scenes the filter would be almost not used at all and where there is low motion or still pictures there would be strong temporal filtering.

I am not a programmer and also new to avisynth so this filter may already exist. Anybody is free to use "my idea" to make a new filter if you want. Perhaps there is methods to do this already?

Just an idea.

Marc FD
14th September 2002, 09:25
I really don't see how you'll gain quality against a Temporalsoftener ?

ronnylov
14th September 2002, 10:29
Let's say I have a clip with still parts like this:

A B C1 C2 C3 D E F G H1 H2 H3 H4 H5 I J

Where C1 - C3 are identical frames and H1 - H5 are other identical frames. On frames A and B I don't want any temporal filters. On frame C2 I would like to apply temporalsoften(1,255,255) to average frames C1, C2 and C3. Then I copy the filtered C2 to the new clip C1, C2 and C3. Frames D, E, F and G I don't want any filter. On frame H3 I apply temporalsoften(2,255,255) to average H1 - H5 and then I copy this averaged frame to the new clip H1, H2, H3, H4 and H5.

I would gain less ghosting on frames A, B , C1, C3, D, E, F, G, H1, H5, I and J compared to a temporalsoften(1,x,y). I would gain less noise on the frame H3. Maybe the overall gain would not be so much but still this was only an idea I had...

OUTPinged_
15th September 2002, 00:13
pretty uneffective.

The thing that will work is motion compensated temporal smoother. Too bad it is a complex thing to do :-(