On the Design of Training and Testing Data for Neural Networks in Image Prediction

The use of neural networks as a nonlinear predictor in many applications including predictive image coding has been successfully presented by many researchers. However, almost all of the research papers have focused on the architecture of the neural network and very little attention has been given to the design of the training and testing data. This paper demonstrates how the choice of the training data could dramatically affects the performance of the neural networks in image prediction. The important design factors of the training and testing data are assessed and the outcomes of the various simulations are presented.