In developing algorithms for art one must create algorithms for color. This isn't so difficult for a naturalistic scene; one just uses colors resembling those seen in nature. For abstract art it is more complicated. One gets into questions of "What color pairs and triplets 'go together?'" and "What color sequences are attractive or unattractive?" Much has been written about this, and the basic rules are systematized in the color wheel. Whether two particular colors "go together" undoubtedly varies from one person to another, since in addition to the basic color and pattern recognition of eye and brain, each individual has associations with various colors and color combinations. If some person we didn't like wore green and blue on some unhappy occasion, perhaps we subconsciously reject that combination.
The aim here was to create algorithmic abstract images which are attractive to the eye. It is well known to abstract artists that color and pattern are exceedingly important for the impression created in the viewer.
1. While abstract art is often colored with only some finite number N of colors, the view taken here is that continuous distributions of colors are to be used. It was also assumed from the outset that random chance would be used in generating the colors.
2. The experiments were done using regions with vertical and horizontal boundaries. There is a particular area of the brain which looks for vertical and horizontal edges and they are a mostly unconscious part of visual perception.
3. It is my view that an interesting image needs both order and randomness. What balance to make between these two, and just how you would measure it, remains unclear. The experiments turned up mathematically interesting patterns which were uninteresting visual spectacles -- and were rejected.
4. A successful color scheme is not independent of the shapes of the colored regions.
5. The image should have the "fractal" property of having interesting features at all length scales, i.e., it should make an impression from 30 ft away, and still reveal interesting detail as the viewer approaches to 3 ft away. Another way of stating this is that it should have both big and small features.
6. Some might think that "random color" is silly and reject the idea categorically. The art here lies in constrained randomness -- how do you constrain the randomness in such a way that the resulting image is "interesting"? This is where the art lies -- in algorithm design.
7. What is a "good" or "bad" image? Is this Art or merely "art"? Given the many subconscious features of human visual perception the only gauge here is the impact upon the viewer. Given two images made by the same constrained-random algorithm, most people will have a definite preference for one of them (not always the same one). Part of this arises from the fundamentals of vision, part from cultural factors, and part from the remembered color experiences of a given individual.
The book "Vision and Art" by Margaret Livingstone has had a considerable influence upon my thinking.