Performance assessment of five neural networks and architecture design for image vector quantization

Various neural network algorithms have previously been introduced to implement vector quantisation for image compression. These include competitive learning VQ, a self-organizing feature map, frequency sensitive learning, an LBG neural network and general learning VQ etc. The first three neural networks are members of the LVQ family. The present paper presents a performance assessment based on experimental results for the above five typical neural networks. The second part of the paper contributes to computing an architecture design for a batch mode GLVQ algorithm which shows the best potential for further development