The edgeflow filter works as follows:
first a sobel filter is applied. The “lightness” of the resulting pixel is examined, where lightness is defined as (max+min)/2. Max and min are max(R,G,B) and min(R,G,B).

The lightness is then compared to a threshold value (set by the user from 0 – 255). If the lightness is > threshold, then the HSL (hue, saturation, lightness) values of the original pixel are computed. The two lightness values are added using a scaling factor for the sobel pixel. Finally the HSL values are converted back to RGB space to give the result.

Optionally, pixels where the sobel value is below the threshold can be darkened, to provide a heightened contrast effect.

The filter can be downloaded here:


The cluster plugin is an implementation of the popular and relatively simple k-means clustering algorithm, where regions of an image are identified by their proximity in color and space to cluster centers and assigned the average color of the region. The cluster centers are updated each frame with the average location of all the points in the region. If distance is given the most weight it looks like a voronoi diagram.

Demonstration video here made in gephex:

Source and linux i386 .so here: