An evolutionary approach to vector quantizer design

Vector quantization is a lossy coding technique for encoding a set of vectors from different sources such as image and speech. The design of vector quantizers that yields the lowest distortion is one of the most challenging problems in the field of source coding. However, this problem is known to be difficult (Gersho and Gray, 1992). The conventional solution technique works through a process of iterative refinements which yield only locally optimal results. We design and evaluate three versions of genetic algorithms for computing vector quantizers. Our preliminary study with Gaussian-Markov sources showed that the genetic approach outperforms the conventional technique in most cases