Application of genetic multi-step search to unsupervised design of morphological filters for noise removal

Effectiveness of extrapolation-directed multistep search in unsupervised design for morphological filters for noise removal

“…dMSXF can be constructed by introducing a problem-specific neighborhood structure and a distance measure and it has been shown to perform well for solving combinatorial optimization problems, particularly on problems for which the landscape is an AR(1) [3] (see Appendix). In our previous work, the landscape of the objective function used for the design of morphological filter was experimentally shown to be similar to AR(1) landscape, and we showed that dMSXF could design more effective filters than that of a conventional crossover [8].…”

“…In our previous work, GA was applied to an unsupervised design of morphological filters [7] based on the opening operation for noise removal in texture images, and we adopted dMSXF as a crossover method [8]. Opening is a typical morphological operation, which composes the resultant image object by arranging a structuring element (SE) inside a target object and removes residual regions that are too small to hold the SE inside.…”

“…Our approach is an unsupervised design; the optimal SEs are estimated only from corrupted images by using the size distribution function [9] as the objective function. Since the feature of an AR(1) landscape was observed in the size distribution function, dMSXF was shown to work very well on estimating SEs [8]. However, dMSXF is an interpolation search method that does not work effectively when parents' characteristics are extremely similar to each other, and an extrapolation search method that explores outside the distribution of the population is required.…”

Effectiveness of Multi-step Crossover Fusions in genetic programming

“…The main difference of these procedures is that the MSXF determines the transition in the local search by Metropolis criterion while dMSXF advances the local search in a deterministic rule. MSXF and dMSXF are defined in a problem-independent manner, and have been successfully applied to various combinatorial optimization problems [10], [11], [12] since the incorporation of local searches into GAs is essential in order to adjust the structural details of solutions [13].…”