Research On No-Reference Quality Assessment Algorithm For Network Video

With the development of information and network technology,network video has become the most important medium.At the same time,the quality of network video varies due to factors such as video shooting,compression,and network transmission.In order to help video operators and network operators objectively analyze video quality and improve users’ viewing experience,we conduct a study on network video quality assessment algorithms.Considering that most of the network videos do not have reference videos,in this paper,we focus on the no-reference network video quality assessment algorithm,including UGC video quality assessment algorithms that do not take network factors into account and network video quality assessment algorithms that take network factors into account.In addition,we design and develop a platform for network video quality assessment.The main work and innovations of the paper are as follows:First,to address the problems of the existing no-reference video quality assessment algorithm with expensive computation and time consumption,we propose a no-reference video quality assessment method in the compression domain.The algorithm proposed in this paper adopts a network structure of spatiotemporal feature extraction+regression prediction.To reduce the computation complexity and time consumption of the algorithm,the spatio-temporal feature extraction of the algorithm is performed in the compressed domain,avoiding the additional computation of video decoding and reconstruction.Specifically,the algorithm extracts spatial features only from I-frames in the video bitstreams,and extracts temporal features from the macroblock and motion vector carried in the video bitstreams.To improve the accuracy of the algorithm,we use support vector regression and random grid search method to predict video quality.The results on KoNViD-lk,LIVE-VQC,and YouTube-UGC datasets show that the method can effectively reduce the assessment time while ensuring the model accuracy.Second,to address the problems of the difficulty in network video quality assessment caused by complex network transmission,we propose a network video quality assessment method based on recurrent neural network,which can assess the overall video playback quality by the past network video clips playback situations,and to reflect the impact of network condition on video quality.The method first extracts multiple features of video clips based on network video streaming characteristics to improve the richness of video features,and then uses recurrent neural networks to capture the semantic correlation between video clips,and introduces a pooling strategy based on the lag effect of human eyes in the neural network to obtain more temporal correlation between video clips.The test results on the SQOE-Ⅲ dataset demonstrate the effectiveness of the algorithm.Finally,based on the above research,we design and develop a video quality assessment platform with B/S architecture,which can evaluate the quality of compressed videos uploaded by users or videos through the network.The platform is divided into two parts:front-end and back-end,where the front-end provides users with compressed video upload interfaces and the back-end is responsible for computing video quality score.The platform provides a reference solution for the implementation of video quality assessment algorithms.