Efficient Local Search in Imaging Optimization Problems with the Hybrid Evolutionary Algorithm

摘要

The paper focuses on efficiency problems related to local search in the hybrid evolutionary algorithm, with application to optimization problems arising in electronic imaging. A two-phase cyclic local search is proposed, which helps defy the efficiency problems. First, random search is performed in the neighborhood of the best chromosomes, with the radius of the neighborhood and the number of the selected neighbors computed as linear functions of time. During the second phase, local search is refined using the cyclic version of the Downhill simplex method (DSM). Alternating random and DSM search re-defines the configuration of the DSM simplex, thus preventing it from converging to a sub-optimal solution. The algorithm utilizes a novel model of image local response, which reduces the overall computational cost of the additional fitness evaluations during the local DSM search. The results of the computational experiments with 2-dimensional grayscale images provide the experimental support for the proposed approach, and clearly indicate that the proposed response-enhanced algorithm of the local DSM search significantly reduces the overall computational cost, without the loss of the quality of the final solution.