ABSTRACT
In this chapter we address the question of how light can be transformed in an efficient way and by finite material into an electrical signal and how the image formed by an array of such transducers can be further processed to achieve a coherent representation of the dynamic visual input. The study of biological systems gives insights about design principles at various levels. At the biochemical level, mechanisms such as enzymatic gain modulation overcome optimizational problems arising from the limited capacity of the transducing material. Neurophysiological and anatomical explorations of living structures, from the simplest to the most complicated, enables one to devise specialized architectures for information processing. Mathematical characterization of these interactions leads to the study of arbitrarily large number of units to achieve a higher-level description of neural networks. The dynamics of the global network, built around local and relatively simple interactions, can show very complicated behavior, such as standing waves, traveling waves, absolutely stable equilibria, resonance, and chaos. While computers let us test cognitive theories by simulations, research in neural networks teaches us how to design computers suited to implement cognitive tasks such as vision.
