3x Speed-up

[w-m]

• Use 1D Gaussian filters, apply twice, along each axis. This is what scipy's gaussian_filter does, and is much faster, yielding the same result
• correctly set mu in relation to gaussian_kernel_width; not always 5
• use -= where appropriate

These changes result in a ~3x performance improvement (on my machine) without any changes to the API or the results.

[jterrace]

This looks cool! Quick question: did you verify that you get the same results with the two sets of test images in the repo?

[w-m]

I made sure that he results are exactly the same for the two pairs of test images of the repo - when called without further parameters.

Since I fixed what I believe to be a bug in the kernel creation, result values will differ if gaussian_kernel_width is set to something other than 11.

[jterrace]

The results are the same on the test images when I tried it and it runs in half the time. Nice!

[jterrace]

Actually, would you mind adding yourself to the AUTHORS file?