Abstract

A new analysis shows how efficient representation structures for higher-level visual tasks, such as segmentation, grouping, transparency, depth perception, and size perception, can self-organize. Here, three simple adaptive algorithms are combined in parallel in a neural-network simulation. A variant of a Hebbian excitatory learning rule provides each neuron with selectivity for patterns. A Weber-Law neuron-growth rule permits the network to learn and classify patterns despite variations in their spatial scale. A new anti- Hebbian inhibitory learning rule permits superposition of multiple simultaneous neural activations (multiple winners), under strictly regulated circumstances, instead of forcing winner-take-all pattern classifications. In the case of stereo transparency, the inhibitory learning both implements a uniqueness constraint and permits coactivation of cells representing multiple disparities at the same image location. Thus, two or more disparities can be active simultaneously without interference. This behavior is analogous to that of Prazdny's stereo vision algorithm,¹ with the bonus that each binocular point is assigned a unique disparity. In a large implementation, such a network would be able to effectively represent the disparities of a cloud of points at random depths, like human observers and unlike Prazdny's method.

© 1990 Optical Society of America