New simple three-layer neural network for image compression

Recently, multilayer neural networks (NNs) have been used to compress still input pictures.1–8 The first NN to be used for this purpose was described by Cottrell, Munro, and Zipser (CMZ),1 who used a simple (single-structure) fully connected three-layer NN. The CMZ NN has a low computational complexity but only moderate compression performance.*