Well this sucks. It looks like the scant memory on the Raspberry Pi 3B causes problems trying to compile
scipy. Look at what's required—yipes.
So I thought, I'll just find a way of doing this with just
numpy. And that died with a memory error while trying to do a simple array copy of an 120K image file. Grr...
Anyway, does anyone have a good-looking 3x3 convolution kernel matrix for Gaussian blur which looks like a webkit 3x blur or a Gimp 10-pixel blur?
Currently, I'm applying this three times and it kind of looks trashy after the third iteration.
0, 0.5, 0 0.5, -1, 0.5 0, 0.5, 0
With only one iteration, it's barely blurred. The Jimp library I'm using doesn't seem to have a way of dividing all values post-apply with the matrix sum, unfortunately. I tried a 5x5 matrix version but I'd need to play with it more to get it right.
Alright... well this one (applied three times) looks better at least.
0.0625, 0.125, 0.0625 0.125, 0.25, 0.125 0.0625, 0.125, 0.0625