Picture Reconstruction Optimization Using Neural Networks

The multilayer perceptron (MLP), as one of neural network types, is a function of one or more predictors which minimizes the error between the inputs and target variables. In this paper, the network architecture is designed and the optimal parameters are chosen. A novel method is proposed to add polynomial features to help get better results on the accuracy of reconstruction picture. The number of epochs for training is also an important fact for time consuming which has strong relationship with the avoid overtraining algorithm. The accuracy of the reconstruction work and the size of time consuming data are examined by experimental work.