Prediction by back-propagation neural network for lossless image compression

This paper describes a prediction process produced by a back-propagation neural network for lossless image compression. The predictor is designed by supervised training of a back-propagation neural network using actual image pixels, i.e. using a typical sequence of pixel values. The significance of this approach lies in the fact that it can exploit high-order statistics and the nonlinear function existing between pixel values in an image. Results are presented for the prediction error image in terms of mean-square error and first-order entropy, and a discussion on the performance of the algorithm is given.