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

Deterministic multi-step crossover fusion (dMSXF) is a promising crossover method in combinatorial problems, which performs multi-step searches focusing on inheritance of parents´ characteristics using a neighborhood structure and a distance measure. DMSXF shows good search ability in various kinds of combinatorial problem. In this paper, we apply dMSXF to unsupervised design of morphological filters based on the opening operation for noise removal. We estimate suitable structuring elements (SE) for texture images corrupted by impulse noise. Here we fix the shape of SE and optimize pixel values of each element of SE. Through the experiments, it is shown that dMSXF works very well on estimating optimal SEs just like typical combinatorial problems.