Code book optimization with a genetic algorithm for vector quantization

We constructed a code book (CB) for vector quantization (VQ) of an image using a real-coded genetic algorithm (GA). Simulated binary crossover (SBX) and a minimum generation gap (MGG) were employed in the GA as a crossover and selection, respectively. We compared the performance three algorithms for construction of a CB: fuzzy c means, read-coded GA, and a combination of two algorithms in which a real-coded GA is used to determine initial code vectors and then the learning vectors are clustered by an FCM algorithm using the initial code vectors. Prototype vectors of FCM clustering become code vectors. Results of experiments showed that there is no notable in performance of the algorithms.