Research On A Neural-network-based Network Video Quality Evaluation Algorithm

With the development of network technologies,online video services have become more and more popular,so the quality of network video has also received more and more attention.The quality of network video can be affected by many factors.For example,video needs to be compressed and encoded before transmission,and network conditions such as packet loss and jitter may occur during transmission,which will have a certain impact on the quality of the network video.Therefore,this paper proposes a network video quality assessment algorithm based on neural network,focusing on network video compression and transmission impairments.The main work is as follows:This paper researches and analyzes the existing network video quality assessment methods,classifies and introduces them,and summarizes the characteristics of commonly used algorithms.Then,the paper proposes a strategy for network video quality assessment using collaborative analysis of compression impairments and transmission impairments.The paper firstly analyzes the damage of compression coding to network video quality,and finds that the quantization parameter reflects the distortion in the video compression,and the ambiguity reflects the low bit rate of the video and the distortion in the process of discrete cosine transform,and the number of jumping macro blocks reflects the distortion in the inter-frame prediction process.So,three characteristic parameters for evaluating the compression damage are proposed:quantization parameter,ambiguity and the number of jumped macroblocks,and it is proved that these three parameters have a good correlation with the network video quality.Then,the paper analyzes the impact of network conditions such as packet loss and delay jitter on network video quality during network transmission,and concludes that the network video quality will be affected in space domain and time domain.The spatial damage is manifested as strain and block effects,while the time-domain damage is represented as stalling events in video playback.This paper proposes six feature parameters that can be extracted from video frame images for spatial damage and time-domain damage:strain degree,block effect,aggregation block effect degree,initial buffering time,stalling average time and stalling frequency.The degree of strain,the degree of block effect,and the degree of aggregation block effect reflect the spatial damage of the video,while initial buffering time,stalling average time and stalling frequency response time-domain video damage.In the end,this paper selects three compression damage feature parameters and six transmission damage feature parameters for the network video quality evaluation.Finally,the paper designs and implements a specific network video quality assessment system based on the nine characteristic parameters that affect the network video quality.The system takes the result of characteristic engineering of nine characteristic parameters of network video as input and takes the subjective quality score of video as the training output.Then the system uses BP neural network algorithm to establish mapping relationship,and finally obtains the objective evaluation quality of network video.The results show that the compressive damage and transmission damage parameters used in this paper can be used to obtain better evaluation results using neural network analysis model,and the system has the characteristics of fewer characteristic parameters and simple operation.Based on the study of network video quality,this paper discusses the impact of compression coding and network status on network video quality,and builds a real-time and effective network video evaluation model by using neural network.These researches provide technical support for the improvement of compression coding,the monitoring of communication quality,the optimization of media services and online video services.