Image Vector Quantization by Neural Networks

Abstract

Image compression constitutes an essential tool for applications such as broadcast television, remote sensing via satellite, teleconferencing, computer communication and facsimile transmission. Compression techniques become necessary to reduce the amount of data to describe a still image or an image sequence correctly. Vector Quantization (VQ) is already known as a very efficient compression method when used in image coding scheme. In fact, using Vector Quantization in place of Scalar Quantization it is possible to reduce the bit-rate of a data compression system at the same quality or viceversa to improve the reconstructed quality image at the same bit-rate. The neural network paradigm represents, for VQ problems, an interesting alternative to traditional algorithms. Neural Networks provide good performance both in quality image reproduction and in computational effort, allowing the use of adaptation schemes to follow the statistics of the incoming images during the coding process. In this chapter, Vector Quantization and Adaptive Vector Quantization applications, solved by using neural net approach, will be presented. Section 2, after a brief review on the theory of Vector Quantization, presents the application of neural network algorithms for two particular cases: colormap design and interframe coding scheme for videoconference sequences Section 3 proposes a solution for adaptive vector quantization problem with a neural network method. Conclusions are given in Section 4.