A reduced reference video quality metric for AVC/H.264 is proposed. This reduced reference metric evaluates a set of features such as blur or blocking and combines these measurements into one quality information using multivariate data analysis. The metric needs a very low overhead in terms of additional bit rate and is verified with data from accurate subjective testing. It does not only outperform standard PSNR but also shows to be superior compared to two full reference video quality metrics. In addition a method for designing video quality metrics is presented. This method is based on multivariate data analysis, a tool that is widely used in chemo metrics and food science to predict latent variables such as taste by evaluating a set of variables that can easily be measured.
展开▼
