Performing Vector Quantization Using Reduced Data Representation

Abstract

We propose a method to improve the performance of vector quantization by using different resolutions of the dataset for each GLA iteration. We discuss the use of wavelet decomposition, principal components analysis and other data and dimensionality reduction techniques on the dataset at different stages of vector quantization. Experimental results on both real and simulated datasets show that the proposed technique outperforms ordinary vector quantization in terms of mean squared error or running time.