Two-Way Data Analysis

Abstract

In the design of video quality metrics, so far mostly (linear) two-way data analysis methods have been applied to two-way feature arrays, generated from the three-way feature arrays by temporal pooling. In this chapter an overview of these methods is provided by discussing different two-way data analysis methods and their corresponding regression models: the multiple linear regression (MLR), the general notation of component models, the principal component regression (PCR) and finally the partial least squares regression (PLSR). For each method, the underlying optimisation criteria are discussed and the resulting advantages and disadvantages compared to other two-way methods are highlighted, but also the different algorithms to obtain these regression models are presented. Additionally, the temporal pooling that is necessary to prepare the features for the application of these methods is discussed in detail.