Fast codebook search algorithm for unconstrained vectorquantisation

Vector quantisation (VQ) is a well known data compression technique which maps an ordered set of real numbers into a single integer. However, it is difficult to achieve accurate compression with the use of unconstrained Voronoi VQ when the codebook level and vector dimensionality are very large, due to the extremely high real-time computational complexity involved in full codebook search. To overcome this difficulty, a classified pre-selection method is proposed. Compared to the conventional full search method, the algorithm reduces the computational complexity involved in the code vector selection procedure by 70%~90% with almost no loss in coder performance, at the cost of only a slight increase in the storage requirement