Random measure fields and the integration of visual information

The fundamental role that a class of layered probabilistic structures may play in the solution of certain complex approximation problems that appear in the integration of visual information, specifically, those that involve the reconstruction of piecewise smooth functions using information from different channels, is explained. Specific models for these structures in the form of Markovian random fields of probability measures are developed, and practical algorithms are given for their computation. Their implementation is discussed both in terms of analog and digital networks (cellular automata). A simple application of this scheme to the reconstruction of piecewise smooth surfaces is discussed in detail. Its application to the solution of more complex problems that involve the interaction of several computational modules, such as the reconstruction of visible surfaces and automatic learning, is outlined as well.<
>