Content-Related Features for Video Quality Assessment Based on Bit Streams

Quality assessment plays an important role in video business. The bit-stream-layer quality assessment model becomes popular in recent years because of its low-complexity advantage. This paper defines and analyzes some content-related bit-stream features derived from encoded MPEG-2 HD videos. We divide the features into three categories: spatial features, temporal features and compression settings. 9 HD videos with different types of content were used in our experiment. They were encoded at 6 kinds of bit rates. Subjective test and feature extraction were performed on all the bit-rate samples. Then we analyzed all the data by correlation matrix. The results show some compression setting features have very high linear correlation (up to 89.89%) with subjective perceived scores. At the same time, some spatial and temporal features are graphed as functions of bit rate to show their ability of depicting spatial-temporal characteristics of videos. In future, these features will be used for building a bit-stream-layer quality assessment model.