SK_Processing footage for a stylized look with SmoothKit

SmoothKit Tutorials: Processing footage for a stylized look with SmoothKit.

 

(c) 2002-10, RE:Vision Effects

 

We present an example of processing footage with SmoothKit so that it has smoother gradients and is less noisy. In this case, we want smoother color transitions so that we can posterize (quantize) the colors in the image for a cartoon look. To accentuate the cartoon look, we wish to draw in edges of the image using edge-finding filters ("Find Edges" or similar filters) which help delineate the important features. A typical problem associated with applying color quantization and edge-finding filters is flickering boundaries and lines due to splotchy areas and noise (variation) in the imagery. This is a perfect job for the Diffusion filter of the SmoothKit toolset, which allows us to preserve the imporant edge boundaries while smoothing the colors in the imagery.

This tutorial may seem rather specific to creating a stylized look, but processing footage in this way can be useful for many tracking and keying jobs too.

The process here uses the simple following process:

  1. Smooth the imagery using SmoothKit Diffusion, so that the next 2 steps in this process are well-behaved over time.
  2. posterize to 4 (or 8) levels and remap the levels (using a color curve) and apply over source with a Luminosity transfer function.
  3. Create an image of object edges using an edge-finding filter (a greyscale image) and multiply (transfer function) the edge image with posterized image

On the left is a source image and on the right is the an image processed with SmoothKit. To smooth the image we have selected the SmoothKit Diffuse filter (RGB mode, Average filter with a large number of iterations). At each pixel, the Diffusion filter uses a small filter size and only filters in pixels if they are "enough similar" to the pixel being filtered... "enough similar" is defined by the user with the "Maximum Deviation" setting of the Diffusion filter, which in this case is set to 10%. The Diffusion filter preserves edges because it only filters pixels into the result if they are enough similar (and it is assumed that at an edge we have pixels that can be sufficiently differentiated by setting the Maximum Deviation tolerance). The diffusion filter creates regions of smooth gradients because by using a large number of iterations the smoothing process grows so that only regions of similar color are filtered and smoothed.

 

 

Below we see that the smoothing allows to create an image with much more stable object edges. Note how some small soft details are removed (picture on the left is not presmoothed).  

 

Now look at the final result without (left) and with presmoothing

 

 

 

  

Here is a motion example with 4 levels of posterization