Feature grouping and figure/ground discrimination: a recursive neural network approach

The authors cast the feature grouping and figure/ground discrimination problems into a combinatorial optimization problem. The cost function for which a global minimum is to be sought is of the same mathematical structure as the energy function of a spin-glass system or of a recursive neural network. Hence, the global minimization problem can be solved by mean field annealing (MFA). The experimental results obtained with an MFA asynchronous implementation show that the method advocated is well suited to solve the problem