Improving the Prediction Accuracy of Objective Video Quality Evaluation

Three different approaches for improvement of objective video quality evaluation are presented in this paper. Improvement is obtained through quality guided temporal pooling, information content weighted temporal pooling, and multiscale analysis. The analysis was performed using five objective video quality assessment measures on two publicly available datasets with subjective quality scores. Only the videos with H.264, H.265, and MPEG-2 types of compression from two datasets were considered. The level of agreement between the subjective and objective quality scores are given through the Spearman rank-order correlation coefficients on complete datasets and subsets of video sequences with the same type of compression. Obtained results show that the performance of objective measures is dependent on the choice of the dataset. The greatest improvement is given by multiscale analysis.