As fred pointed out, Voroni is a sparse color method, and while related to a Morphology Distance Gradient, it is not the same.
It is also a 'point' solution where it looks for the nearest colored point that was given in a list. The provided image only provides a 'base canvas' on which it works, and which it will overwrite (unless a write or clipping mask was set).
You can however convert the edges of an image into a 'point list' for voronoi point filling!
See the 'shepards fill' example in...
http://www.imagemagick.org/Usage/canvas/#sparse_fill
The key is to only generate a list of edge pixels so that the 'point list' is kept to a minimum.
(This is something I would have liked to build into IM but never got around to it)
I did make a note morphology distance function (which their is half a dozen distance metric for) can be used to generate a 'fast voronoi' fill method. however this requires that while the distance function is generated in one channel (typically a negated alpha channel), the values in the other channels (color) of any newly discovered the 'nearest pixel' also need to be copied at the same time.
It is a tricky solution but works without needing a point list, and for any number of pixels, is just two passes through the image just as the distance function does, instead of the current one pass with a list. Also you can make selection
of different and simpler distance metrics (like a Knights Move, or Octagon) than just Eucliedien.
It is part of first stage in a true 'diffusion hole filling' technique, which understands boundary conditions, unlike a 'shepards fill' technique that is already available, and even provides a faster blurred image stack alternative.
It is a variation that requires some addition programming which I have not had the time to make.
I have however attempted tried to implement in other ways as quick experiments (in hole filling), quite some time ago, but nothing really useful came from it.