Macroblock level quality assessment using video-independent classification

In this paper we propose a no-reference objective quality assessment of compressed video at a macroblock level. We propose a video-independent approach to quality assessment of individual macroblocks. Features are extracted from video bit steams and reconstructed videos. The feature variables are validated through the use of stepwise regression. The classification model is generated using a reduced-model polynomial networks and SVMs. For higher accuracy of quality assessment, the paper proposes the use of multinomial logistic regression to report the probabilities with which the macroblock class label is assigned. With the use of six video sequences, the experimental results show that video-independent classification at a macroblock level is permissible with a classification accuracy close to 89%.