Databases Reference
In-Depth Information
10
Vector Quantization
10.1 Overview
By grouping source outputs together and encoding them as a single block, we
can obtain efficient lossy as well as lossless compression algorithms. Many of
the lossless compression algorithms that we looked at took advantage of this
fact. We can do the same with quantization. In this chapter, several quantization
techniques that operate on blocks of data are described. We can view these
blocks as vectors, hence the name “vector quantization.” We will describe several different
approaches to vector quantization. We will explore how to design vector quantizers and how
these quantizers can be used for compression.
10.2 Introduction
In the last chapter, we looked at different ways of quantizing the output of a source. In all
cases the quantizer inputs were scalar values, and each quantizer codeword represented a single
sample of the source output. In Chapter 2 we saw that, by taking longer and longer sequences of
input samples, it is possible to extract the structure in the source coder output. In Chapter 4 we
saw that, even when the input is random, encoding sequences of samples instead of encoding
individual samples separately provides a more efficient code. Encoding sequences of samples
is more advantageous in the lossy compression framework as well. By “advantageous” we
mean a lower distortion for a given rate, or a lower rate for a given distortion. As in the previous
chapter, by “rate” we mean the average number of bits per input sample, and the measures of
distortion will generally be the mean squared error and the signal-to-noise ratio.
