Relaxation labeling using Lagrange-Hopfield method

Relaxation labeling (RL) is a class of parallel iterative numerical procedures which use contextual constraints to reduce ambiguities in image analysis. A novel RL algorithm is proposed. RL is posed as a constrained optimization problem and the solution is found by using the Lagrangian multiplier method and a technique used in the graded Hopfield neural network. In terms of the optimized objective value, the algorithm performs almost as well as simulated annealing, as shown by the experimental results. Also, the resulting algorithm is fully distributive and is suitable for analog implementation