Research On Temporal Pooling For Video Quality Assessment

With the development of multimedia technologies,video applications have been used in every aspect of our daily life.In order to guarantee the quality of a video service,it is necessary to evaluate the video quality.The objective video quality evaluation plays an important role in quality evaluation due to its real-time and low-cost advantages.Acting as a key part of objective video quality evaluation,the performance of temporal pooling affects that of the whole evaluation system.In this thesis,temporal pooling for video quality assessment is addressed.Temporal pooling integrates the series of frame quality scores into one quality score for the entire video sequence using temporal information,given the quality of each frame.This thesis evaluates the relationship between the quality of frames and the quality of the video sequence with constant frame quality,where temporal complexity has a strong influence on video quality.Then an effective metric of temporal complexity is determined according to the experimental results.Finally,a temporal pooling model is built to evaluate the video quality with constant frame quality.In practice,not all video is with constant frame quality.Owing to channel errors and the other factors,there may occur several successive frames with low quality.Concerning the video with successive low quality frames,research has been conducted on the influence of quality reduction and the number of frames with low quality on the overall video quality.It is found that the perception for a long video is not always reliable,and thus it is favorable to integrate video segments rather than to integrate frames to form the video sequence.Further research is carried out to find the influence of quality reduction,the number of frames with low quality,and the distribution of the frames with low quality on the quality of video segments.Through analysis of the experimental results,the relationship between the three factors mentioned above and the video segment quality is determined,and then a temporal pooling model for video segment is established.Finally,based on GOF(Group of Frames)concept,a temporal pooling algorithm is proposed for videos with successive low quality frames.The experimental results have shown that the proposed temporal pooling algorithm gives good consistency between subjective and objective scores.