nicholasudell / normalizr Goto Github PK

This could be a lot faster.

Currently each pixel is read individually and then written individually.

Using GetPixels to grab a whole rectangle and then write a whole rectangle would speed this up a huge amount.

Some of them could work in a generic sort of "apply convolution filter to this image" approach, but some (such as phase congruency for sharp edge detection, e.g.) would not.

Example filters needed:

• Sobel
• Prewitt
• Phase Congruency

Currently if a non-transparent pixel touches the border of the image, it's set to hard blue. It needs something a little closer to a real normal map here.

Perhaps using 2 2x1 convolution filters instead of 2 3x1 filters?