Image data compression using a self-organizing neural network with adaptive thresholds

Abstract

The authors describe a novel neural network architecture which, when applied to image data compression, is capable of real-time, one shot, vector quantization. The major feature of this system is the utilization of an adaptive threshold mechanism. The architecture comprises a two-layer network and a winner-take-all cluster with adaptive thresholds. This system was used to generate a codebook for digital images.