Autonomous real time learning in neural networks for image compression

Vector quantisation (VQ) is a technique applicable to the compression of still and motion video images. The efficiency of the compression scheme depends on how well the statistical properties inherent in the image are extracted and used. Neural networks because of their fast parallel search capabilities make good vector quantisers. A neural network used for VQ must be able to learn the local statistical properties of the image very quickly if it is to process video images efficiently in real time. However, Quick learning is not a property normally found in neural networks developed to date. The paper describes a self organising neural network VQ architecture that has a fast autonomous learning algorithm suitable for use in real time image compression