Performance Comparison

Abstract

In this chapter, the different data analysis methods discussed in this book are compared by using two example video quality metrics. Following an introduction of the performance metrics used to determine the individual video quality metric’s performance, the selection process of the data sets used to build the example metrics and their properties are outlined. As the data used for training and validation plays an important role in a data driven approach, great care has to be taken that the basis of the models, the subjective quality ratings and the corresponding video sequences, are selected appropriately. Finally, the performance of the data analysis methods is compared in two parts. In the first part, the different data analysis methods, MLR, PCR, PLSR, unfolding and bilinear PLSR, 2D-PCR and trilinear PLSR are compared to each other, whereas in the second part the best data analysis method and corresponding metric is compared to selected state-of-the-art video quality metrics, before concluding with a short summary and general assessment of the data driven design approach.