A novel no-reference video quality assessment algorithm

Abstract

The no-reference video quality evaluation method has become a hotspot and difficulty for video quality evaluation research due to its convenience and lack of reference information. This paper proposed a spatio-temporal domain combined no-reference video quality assessment method. On the basis of no reference image quality evaluation, temporal domain information is added, and multi-layer perceptron neural network(MLPNN) is used for training to obtain a no-reference video quality assessment model. The paper focuses on the extraction and pooling of temporal domain parameters. Standard deviation(STD) and GGD scale parameters extracted from the matching frame difference (MFD) are used as temporal domain feature parameters. Mean pooling of absolute value of differences is adopted to pool temporal domain feature parameters. The proposed framework is trained and tested on the LIVE (Laboratory for Image and Video Engineering) video quality database. The results have demonstrated the proposed method performs well in predicting video quality and it is current state-of-the-art algorithms.