Perceptual grouping as energy minimization
P. Kahn, A. Winkler, Chee-Yee Chong · 2002
A method that uses simulated annealing to perform perceptual grouping is introduced. The first step in the technique maps the perceptual grouping problem to an energy function in which the minimum energy corresponds to the preferred perceptual groups. The second step optimizes this perceptual grouping energy function in order to find the best perceptual groups. By identifying and generalizing the role and composition of group constraints in the perceptual grouping process, this approach explicitly incorporates the gestalt sense of whole, which has not been suitably incorporated in previous techniques. In part, this is achieved by using multiple group constraints that may be simultaneously applied and evaluated. The approach is highly modular, since it allows simple and separate modification of global group evaluation constraints, group permutation rules, and the annealing schedule. Results for perceptual grouping problems of increasing difficulty are presented.>