No-Reference Perceptual Video Quality Measurement for High Definition Videos Based on an Artificial Neural Network

In this paper, we present a novel no-reference (NR) model for perceptual video quality assessment, which can make quality prediction for high definition (HD) videos. This model is based on an artificial neural network (ANN) implemented by the back-propagation algorithm (BP), named as BP-ANN. Six video features are extracted from temporal and spatial domains as the input vectors. Subjective assessments are carried out by using double stimulus continuous quality scales (DSCQS) as the mean opinion scores (MOS), which are desired responses to the output layer. We establish a sample database to store all the videos, feature vectors and its corresponding MOS. Due to the combination of chrome features incorporated with a good use of regions of interest (ROI), our model can achieve good performance for the video quality prediction.

I. Introduction

Advances in video compression technologies and communication systems enable us to obtain high-quality videos much more quickly and conveniently than before. Moreover, the prevalence of digital multimedia has brought increasing interests in introducing video quality issues into a variety of applications, especially for television industries. Television videos are produced for the enjoyment or education of human viewers, and thus it is the viewers' opinion of the video quality that is important.