Automatic Evaluation Methods Of Image Quality For HEVC Streams Based On Coding Parameters

Network video streaming plays an important role in distance education,telemedicine and video surveillance.Higher quality of video transmission becomes more significant for users.However,the resources of the network systems are still relatively limited.To ensure the smoothness of the real-time video play,the bit rate must be reduced as much as possible.However,low bit rate of video will reduce the video quality and user experience.Therefore,it is necessary to transmit high-quality video according to system resources.But the existing video image quality assessment methods can not meet the actual needs.One of the reason is that the current evaluation methods mostly aiming at evaluating the objective loss of video data,and there is a big gap between the evaluation results and the actual observation results of human eyes.Some evaluation methods are consistent with human eye perception results,which can not achieve a balance between the complexity and accuracy of the algorithm.Besides,these methods can not meet the needs of real-time video quality assessment either.Thus,it is necessary to develop an automatic video image quality assessment method that both integrates human visual characteristics and meets real-time requirements.An automatic image quality evaluation method for video compression stream is proposed by using video stream coding parameters and combining visual masking effect.Generally,video quality assessment at video decoder side takes a lot of time to complete decoding of video stream,which can not meet the real-time requirements of video evaluation.This paper proposes to use coding parameters to evaluate the quality of video image.It can evaluate the quality of video image only extracting a few parameters from the stream.After analyzing the degradation factors of HEVC video stream,SSIM is selected as the evaluation index of video stream image quality,and the dependence relationship between SSIM and degradation factors is further analyzed to determine the video stream coding parameters to be extracted.Secondly,the masking effect of human visual effect is used to adjust the video stream image quality assessment,so that the results of video stream image quality assessment are more in line with human visual perception.In order to make use of masking effect and distinguish from the general extraction of frame-level parameters,a new method of extracting parameters using smaller CTU(Coding Tree Unit)in HEVC(High Efficiency Video Coding)is proposed.Texture features of video are represented by the partition depth of quadruple coding tree,and temporal features are represented by the ratio of CU(Coding Unit)of skip mode.Finally,using CTU-level parameter quantization parameters QP(Quantization Parameter),bit rate,CTU partition depth,skip mode use ratio,SSIM value as parameters,using random forest to model the image quality of I frame and B/P frame respectively,and finally realize the evaluation of video stream image quality.The model was tested and analyzed by using three evaluation indicators and CIF format video sequences with the same resolution as the training set and multiple resolution video sequences in HEVC official test video.The experimental results show that the B/P frame model is better than the I frame model.The errors between SSIM value and real value of all test videos are little,and the evaluation result is more accurate.In addition to the low correlation and instability in the case of excessive video resolution,other cases basically show good evaluation results.