Constructive evolution of morphological filters for high-speed embedded image processing

This paper presents a constructive variable length genetic algorithm for the optimisation of morphological filters based on arbitrarily connected sequences of compact look-up table operators. The algorithm employs accretive and compositional evolutionary processes in conjunction with problem specific diversity management and an adaptive memory-based fitness-scaling scheme. The application of the algorithm to standard filters shows that the use of arbitrary connectives results in significantly reduced sequence lengths when compared to a previous disjunctive approach. This improvement has the effect of reducing the processing time and power consumption of a high-speed binary image processor by factors of up to 43%.