Voronoi: unexpected behavior for multi-channel images

[ei14]

Take this image:

Actual output of voronoi:

Expected output:

Per the documentation, the voronoi command should "Compute the discrete Voronoi diagram of non-zero pixels in selected images." This is not happening upon applying the voronoi command to the example image provided.

[dtschump]

Two things:

• Command voronoi works primarily for scalar-valued images (1 channel). When multiple channels are provided, the command is performed channel by channel.
• Non-zero pixels should have different values in those different channels, otherwise they will form different regions with the same colors and thus, we won't be able to see the region boundaries.

So for your input images, which has 4 channels and only 0 or 255 values for each channel, you don't get your expected result.
Worarkound: Transform first your image as an indexed image,

$ gmic input.png +colormap 0 index.. . voronoi.. map.. . rm.

and you get the expected result.

Anyway, I'm not saying this channel-by-channel behavior is desirable, so we can discuss about it and make this command evolve for future G'MIC versions if necessary.