A genetic algorithm design for vector quantization

A new genetic learning algorithm is described in this paper to perform vector quantization for image compression. The algorithm is initialized by generating several random code-books to create a gene pool. A genetic algorithm is then developed and activated to improve on the performance of the initial code-books. After a certain number of generations, a final code-book is selected in accordance with the operation of the genetic learning algorithm. Experiments show that the proposed algorithm outperforms the competitive learning algorithm for vector quantization of digitized images. Significant improvements as high as over 15% have been achieved by the proposed genetic algorithm. In addition, potential for a new direction of vector quantization and image compression research is opened up by considering the establishment of a gene pool and various designs of crossover breeding and mutation operations in the algorithm design.