ABSTRACT
This section is devoted to the applications of neural paradigms on the domain of data and, in particular, still and motion picture compression. Neural networks are inherent adaptive systems, thus they are suitable for handling nonstationaries in image data. Artificial neural networks have been employed with success in at least three approaches to image compression, namely, predictive coding, transform coding, and vector quantization. In predictive coding compression is obtained by exploiting spatial and spatiotemporal redundancy by means of prediction methods, while in transform coding an image or a sequence of images is transformed and the coefficients of the transformation are coded. In vector quantization the data are organized in vectors. The space of vectors which is obtained is then divided into regions and a reproduction vector is calculated for each region. In this section these techniques will be briefly illustrated and for each of them the corresponding neural paradigms will be analyzed. The drawbacks and advantages of the neural techniques with respect to the more traditional ones will be outlined.
