How to evaluate search methods for vector quantization
Erik Agrell, P. Hedelin · Chalmers Publication Library (Chalmers University of Technology) · 1994
Fast search methods for vector quantization are a necessity for benefiting from the performance gains of large-sized codebooks in real-time applications.We take on two 4096 vector quantizer codebooks as illustrative examples for our study.The performance of a set of search procedures is compared.Several aspects of complexity are discussed.We compare average computational complexity and maximum computational complexity in the light of real-time usage.We address storage requirement and the computational complexity required to set up the search procedures.Moreover we illustrate distortion as a function of computational effort.One main conclusion is that an accurate comparison will not yield a single answer.Depending on what aspects that are highlighted in a test, either of the three search procedures in study can be elected as the winner.1.