Nonlinear mechanisms and higher-order statistics in biological vision and electronic image processing: review and perspectives

The classical approach in vision research-the derivation of basically linea r filter models from experiments with simple artificial test stimuli-is cur rently undergoing a major revision. Instead of trying to keep the dirty env ironment out of our clean labs we put it now right into the focus of scient ific exploration. An increasing number of scientists are using natural imag es in their experimental work, and concepts from statistics and information theory are employed for the theoretical modeling of the results. The new a pproach has a close relation to basic engineering strategies for electronic image processing, since its major concept is that biological sensory syste ms exploit the statistical redundancies of the environment by appropriate n eural transformations. The standard engineering methods are not sufficient, however. Even such a basic biological feature as orientation selectivity r equires the consideration of higher-order statistics, like multivariate wav elet statistics, cumulants, or polyspectra. Furthermore, there exists an ab undance of nonlinear phenomena in biological vision, for example the phase invariance of complex cells, cortical gain control, end-stopping, and a var iety of extra-classical receptive field properties. These amount to nonline ar combinations of linear wavelet filter outputs, which are required to exp loit higher-order statistical dependencies, and make it necessary to consid er unconventional modeling approaches like differential geometry or Volterr a-Wiener systems. By use of such methods we cannot only gain a deeper under standing of the adaptation of the visual system to the complex natural envi ronment, but we can also make the biological system an inspiring source for the design of novel strategies in electronic image processing. (C) 2001 SP IE and IS&T.