The Application of Genetic Algorithms to Feature Recognition and Object Counting

1. Introduction Even for the human, visual perception is a complex task, demanding vast resources of prior knowledge and involving the combined computational power of some 10 10 neurons. Yet, in spite of this formidable processing capability, the eye occasionally arrives at erroneous interpretations – as is well-known from the considerable number of optical illusions it is subject to. These illusions are ultimately due to the fact that the eye-brain system takes shortcuts in interpreting visual data, by adopting a hypothesis-based interpretation schema rather than the more obvious 'linear' approach in order to save computation. Quite apart from ambiguities and illusions, visual perception presents serious intrinsic difficulties in many practical situations: take, for example, the needle in the haystack problem, or the robot which has to pick a bracket from a pile of parts. In such cases there is such a welter of lines, curves, holes, edges and other features in the picture that grouping them to form objects is an especially complex matter (again, hypotheses have to be made if efficient means of solving such problems are to be found) [1].