he problem of image compression via neural networks (NNs) is considered. A parallel architecture is presented in which different parts of the image are processed by different NNs according to their complexity. Simulations are presented which show the simplicity and power of this approach.< >
