Evolving color constancy

Evolutionary algorithms are frequently used to find optimal solutions in many different problem areas. They are based on Darwin s principle of survival of the fittest (Darwin 1996 Maynard Smith 1993). A population of individuals has to compete with other individuals for access to food and mates. Only the successful ones are allowed to reproduce. This leads to the reproduction of certain inheritable traits into the next generation. 

The theory of evolution provides answers to many questions pertaining to every day fife. For instance, why are we afraid of certain animals but not of others The simple answer is that some animals are more dangerous than others, i.e. certain types of spiders or snakes are very dangerous. Therefore, it makes sense that many people are afraid of these types of animals. People who like to play with spiders or snakes may have had a slightly lower reproduction rate because some of them died when playing with these animals. 

Evolution basically tells us what life is all about. It can be summed up as survive, eat, reproduce (Dennett 1995). The first and major goal is survival, i.e. not to get eaten by others. The second is finding food. Without food an individual cannot survive. The hunt for food induces a selection pressure on the prey population. If the other two goals are satisfied, i.e. there is no danger of getting eaten by others and the individual is not hungry, then the goal is to find a mate and to reproduce. 

Estimate from current element L,-(x, y) center Estimate from left element L,(x — 1, y) left 

Estimate from element above L,-(jc, y — 1) up Estimate from element below L, (x, y + 1) down 

---

Figure 8.9 Performance of best evolved individual on three different fitness cases. The images in the first row show the reflectance images. The images in the second row show the virtual illuminant. The images in the third row show the input images presented to the individual. The images in the fourth row show the illuminant that was estimated by the evolved individual. In the last row, the estimated reflectance is shown. (Reproduced from Ebner M. 2006 Evolving color constancy. Special issue on evolutionary computer vision and image understanding of pattern recognition letters, Elsevier, 27(11), 1220-1229, by permission from Elsevier.)...

Ebner M 2006 Evolving color constancy. Special Issue on Evolutionary Computer Vision and Image Understanding of Pattern Recognition Letters 27(11), 1220-1229. 

Reproduced from Ebner M. 2006 Evolving color constancy. Special... 

Color constancy algorithms can be derived by looking at the underlying physics of color image formation. An interesting question is whether color constancy algorithms are an acquired capability of the visual system, i.e. if they are learnt, or whether these algorithms are hardwired, i.e. evolved. 

---