Image quality assessment using reduced-reference nonlinear model

This paper presents a reduced-reference nonlinear model driven image quality scheme that is based on a Neural Network statistical estimator, namely Multilayer Perceptrons (MLP) and that is optimized to the Mean Opinion Score (MOS) scale for the combination of input different objective quality measures. In order to examine the performance of the models and identification of how well the model estimates the MOS as a reference model, a linear model is chosen. Also, in this paper is analyzed the effects of an input measures selection on the quality of the MOS estimation.