IMAGE-CODING BY A NEURAL-NET CLASSIFICATION PROCESS

A self-organizing neural network performing learning vector quantizati on (LVQ) to compress image data is proposed. By using unsupervised lea rning, our LVQ neural model approximates optimal pattern clustering fr om training images through a memory adaptation process, and builds a c ompression codebook in the synaptic weight matrix. The neural codebook , trained by example pictures, can be used as a codec to compress and decompress other pictures in a speedy fashion. Special image types, su ch as fingerprints, verify this property in our experiments. Our appro ach is compared with other recently developed neural VQ models (neural gas, growing cell structures, and conscious competitive learning) and methodological premises are discussed. The performance of our model i s also compared with JPEG and wavelet methods. Other advantages of our neural codec model are its low training time, high utilization of neu rons, robust clustering capability, and simple computation. Further, o ur model has some filtering effects through special training methods a nd learning enhancement techniques for obtaining a compact neural code book to yield both high compression and high picture quality.