A neural network approach to image compression
D.S. Erickson, K. S. Thyagarajan · 2003
Discusses some results of the design of a learning vector quantizer for image compression and the effects of the various parameters on the learning convergence. It is shown that good visual quality can be obtained at low bit-rates by using a multistage self-organizing neural network. The self-organizing network architecture is described. Experimental results of using the self-organizing network for codebook generation are described. The learning in the self-organizing network occurred very fast, achieving near maximum learning within a few tens of thousands of iterations.>