This research project concerns a structured class of vector quantizers (VQ) with applications in image and speech source coding. They are called geometric VQ's because they use the geometry in N-dimensional space resulting from equipotentials of the N-th order probability density function. These structured VQ's have greatly reduced implementation complexity compared to general or unstructured VQ's, yet their performance can be comparable in practical image coding applications. As a result the principal investigator can reasonably look at adaptive geometric VQ's with manageable complexity. It is believed that this advancement can be crucial to substantial further improvements in VQ and its application to speech and image coding for the purpose of reduced transmission and storage requirements.
