Negative fuse network

Standard regularization theory combines least squares methods with smoothness constraints, whichleads to quadratic variational functionals with a unique, global minimum (Horn and Schunck [2];Hildreth[3}; Pogglo, Torre and Koch[4}; Poggio, Voorhees and Yuille, [5]; Grimson[6]). Thesequadratic functionals can be mapped onto linear resistive networks, such that the stationary voltagedistribution, corresponding to the state of least power dissipation, is equivalent to the solution ofthe variational functional (Horn[7}; Poggio and Koch, [8]). Data is provided through the correctchoice of data-dependent voltage sources and resistors at each node. Much research has gone intoextending these quadratic variational functionals to allow for discontinuities. Geman and Geman[9]first introduced a class of stochastic algorithms, based on Markov random fields, that explicitlyencode the absence or presence of discontinuities by means of binary variables. Their approach wasextended and modified by numerous researchers to account for discontinuities in depth, texture