The Word Portraits that I have been creating lately use an algorithm that analyzes a starting image and finds rectangular patches of a reasonably consistent color. These are then filled in the generated image with words or letters painted with the average color in the rectangle. I thought it might be interesting to explore piecing the generated image together from fragments that aren't based on words and letters.
The subject for this study is my little four-legged friend Ginger. She's a six year old Cockapoo (Cocker Spaniel / Miniature Poodle cross) , very friendly with kids but a little shy around adults she doesn't know. I don't think she'll mind if you look at a few pictures of her.
The first image just uses the letter 'O' to fill the color-consistent regions. The second uses a leaf-like patch made from bezier curves. They both leave quite a bit of the black background peeking through so the overall image is darker and loses some realism.
These next two images use simple rectangles and circles as the fill shapes. I have also 'overdrawn' them a bit bigger than the original patch which tends to fill up the black gaps. Our eyes pick out the straight edges in the rectangle filled version so it looks less realistic than the one filled with circles. If you squint or look from far away the circle-filled image looks pretty much identical to the original image. This technique could be (and probably has been) used to save images in a compressed form.
This last image uses variations on a spiral shape to fill the color-consistent patches. The area around one of Ginger's eyes is expanded on the right hand side to illustrate the spiral details.