Color devices such as scanners, printers and CRTs have device specific coordinate systems. It is desirable to be able to produce mappings between the coordinate system of a particular device and a device independent coordinate system that closely approximates human perception. In this way, color coordinates can easily be specified, transformed, or transported between various input and output devices. Mappings between color coordinate spaces can be achieved by function restoration, when a number of input-output samples of the mapping are available. Feed-forward multi-layer neural network have been shown to be able to perform non-linear non-parametric functional restoration, as is the case with color coordinate mapping. This type of network was used to map the Lab coordinate space onto the RGB coordinate space of an actual and a computer modeled dye sublimation printer. A neuron activation function, is introduced herein, which has advantages that would be useful to function restoration problems such as color mapping. The effectiveness of this model is tested by observing the error between the model's prediction and the ideal correct output on a number of known samples.