Adaptive learning vector quantizers for image compression

We investigate adaptive vector quantization for image compression based the idea of gold-washing. The technique is a mechanism for testing the usefulness of a code vector in a codebook. It thus provides a tool for developing new ways of creating code vectors dynamically based on the input data. In this paper, we propose a new algorithm to quantize an input for which a close enough code vector can not be found. It guarantees that the compressed result is within pre-set distortion. We also use a learning algorithm to produce new code vectors from useful existing ones.