A data compression system using neural network based architecture

A data-compression system is presented using a neural-network-based architecture that will accept a whole scene of data, compress it for storage or transmission, retrieve it, and reconstruct the original scene. The architecture of the network consists of a feedforward four-layer network. The network is trained by using a nested training algorithm. The authors (Proc. Int. Joint Conf. on Neural Networks, Washington, DC, Jan. 1990) had presented limited results on the performance of this system. A more detailed performance evaluation of this system is presently given using multiple gray-level inputs and variations of training methods