The brush stroke accept/reject algorithm looks at the axis-aligned bounding box that surrounds each candidate brush stroke's affected area. It calculates the cumulative pixel color similarity between the input image and the painting over this area, before and after the candidate brush stroke.
Pixel color similarity is the sum of the absolute difference of each RGB color channel divided by 3.
The candidate brush stroke is accepted and kept by the painting if and only if it represents an improvement in similarity over its affected area AABB.
Thanks for more details. I've been getting into street photography lately and wanted to make renders of the images using old-school image processing techniques. There are a ton of them out there that could mix-n-match for some really cool effects.
Here is some footage of the program for those interested: https://www.youtube.com/watch?v=9iWu4ArbD5Q
Reminded this brilliant video https://youtube.com/watch?v=9tHTCNr5H8s
I'm curious about the greedy algorithm vs something like compressed sensing. Can you expand on the algo choice and how it works?
Sure!
The brush stroke accept/reject algorithm looks at the axis-aligned bounding box that surrounds each candidate brush stroke's affected area. It calculates the cumulative pixel color similarity between the input image and the painting over this area, before and after the candidate brush stroke.
Pixel color similarity is the sum of the absolute difference of each RGB color channel divided by 3.
The candidate brush stroke is accepted and kept by the painting if and only if it represents an improvement in similarity over its affected area AABB.
Thanks for more details. I've been getting into street photography lately and wanted to make renders of the images using old-school image processing techniques. There are a ton of them out there that could mix-n-match for some really cool effects.