Параметризація схем обробки УЗД сонних артерій за допомогою еволюційних алгоритмів.

Abstract

Nowadays, a lot of methods or sequences of image processing algorithms or, in other words, image processing schemes have been designed. Automatic selection of an image segmentation method (scheme) presupposes the existence of some rule which establishes a correspondence between a scheme (processing algorithms sequence) and the values of current image parameters. These rules allow parameterizing methods (schemes). So, there is a problem of determination of such rules. The solution of this problem is formalized and presented in the given paper. The parameterization of segmentation methods has been performed on the basis of an evolutionary algorithm. The algorithm searches for some set of texture parameters and their ranges. It allows us to assign some subspace in the whole space of all the texture parameters to each filtration scheme. Textural parameters are calculated on the basis of gray-level co-occurrence matrices. The proposed algorithm has been implemented and tested. 325 different image processing schemes have been used to test the algorithm. The comparison of different versions of the proposed algorithm has been conducted. The optimal parameters of evolutionary algorithms have been chosen: mutation probability, crossover probability, population size, number of generations, number of simultaneously developing populations, as well as the percentage of genes which are being activated on the initial step of population generation. Testing has been conducted on the basis of texture parameter values of ultrasound images of human carotid arteries.