ABSTRACT
Vector quantization (VQ) can be regarded as a process of pattern classifying and matching. Neural networks, consisting of analog computational elements, can accomplish basic matching and classification of patterns. This represents an alternative to traditional VQ, which may be implemented with high speed but lower resolution analog neural circuits. The self-organizing feature map is an unsupervised neural net based clustering algorithm that can be used to design a VQ. An encoding process with vector quantization can be divided into three steps: convert the analog signal to digital data, find the nearest codevector to the input vector by searching the whole codebook, and code the nearest codevector by the corresponding codeword. The neural network VQ is a new vector quantizer quite useful in encoding still pictures as well as image sequences, based on neural networks. Compression, when applied to images, is usually known as image coding.
