SK_Cleaning up blocky and noisy DV footage

SmoothKit Tutorial 1 : Using SmoothKit on DV footage

(c) 2002, RE:Vision Effects

In this tutorial it is our goal to demonstrate how to use SmoothKit to clean up problematic DV footage. In order to clean up footage, it will often the case that you will need to use several of the filters of SmoothKit in succession. This tutorial attempts to give you a feel for how to work with SmoothKit to achieve the results you desire.

This tutorial also assumes that you've read the SmoothKit manual and are at least a little familiar with how each SmoothKit filters work.

This tutorial addresses two typical DV footage problems. This example has a lot of video noise in the dark areas and some of the object edges are pretty "ugly" as a result of deinterlacing.

Above is an original movie as captured by camera and then deinterlaced in After Effects.

This is a frame from the original material, blown up by a factor of 2.

1) Well, in order to clean up some of the noise you might be tempted to apply a simple Gaussian blur. The result would probably look something like this:

Fig 1. The above movie is the result of applying a simple Gaussian blur to the movie. Note the edges of the hand have been blurred quite a bit.

2) Next, in order to try to retain edges, we modify the SmoothKit Gaussian blur so that pixels can only change by 10.69% (or some other small amount). The noise overall is smoother, but the edges of the hand are still staircased quite a bit.

Fig 2. Application of SmoothKit Gaussian with a small maximum deviation setting.

3) Next, we switch tactics and apply Staircase Suppress instead of the Gaussian filter. Note that the edges are nicer, but there is still quite a bit of noise in the image.

Fig 3. Application of Staircase Suppress by itself.

4) Next we apply both SK Staircase Suppress and SK Gaussian in succession. Note that we have set the Max Deviation % on the Gaussian filter to be small in order to try and retain edges.

Fig 4. Staircase Suppress and Gaussian used in succession.

5) Directional: here we apply the Staircase Suppress and then use the Directional filter. We specify that the Directional filter should use the image itself to determine the directions in which to smooth, so filtering occurs along edges (and not across them).

Fig 5. Application of Staircase Suppress followed by Directional.

6) After application of Staircase Suppress and Directional the arm is very smooth, however we would like to reduce the background noise some more. So we insert Diffusion in between Staircase Suppress and Directional (remember that Diffuse smooths using a small kernel size, and uses neighboring pixels when filtering only if the max deviation threshold is met) . In addition we turn down the Directional length (down to 1 here), because the Diffusion plugin does some smoothing near the edges that we have decided that we don't need to resmooth.

Fig 6. The chain of Staircase Suppress, Diffusion and Directional.

Summary: These filters are made to work together and will often need to be cascaded in different ways, depending on your goal and your content. Finally we show you the original movie and the processed movie.