Vector quantization based on steady-state memetic algorithm

A novel memetic algorithm (MA) for the design of vector quantizers (VQs) is presented in this paper. The algorithm uses steady-state genetic algorithm (GA) for the global search and K-means algorithm for the local improvement. As compared with the usual MA using the generational GA for global search, the proposed MA dramatically reduces the computational time for VQ training. In addition, it attains a near global optimal solution, and its performance is insensitive to the selection of initial codewords. Numerical results show that it can save more than 70% of computation time while maintaining a comparable performance as previous MA.