A Genetic Algorithm Based Method for Determining Two-level Morphological Operators

In two-level training of binary morphological operators, several operators based on different subwindows of a larger window are designed in the first level of training, and then their outcomes are combined in a second level of training. It has been shown that such training scheme yields two-level operators with better performance than the ones designed directly on the large window. A difficult part of the two-level training scheme is, however, to set up the number of first level operators and their respective windows. To avoid the need to specify these parameters, we propose the use of a genetic algorithm based approach. Experimental results show that the proposed method can find twolevel operators with better performance than those manually designed by experienced users. Keywords-morphological operator; image operator training; multilevel training; genetic algorithm.