Research On No-Reference Video Quality Assessment Method

With the development of video coding technology,video applications become more and more extensive,the amount of video also increased.Through a set of video quality evaluation system can timely get the video quality,so that to better provide high-quality video at the front ends.Evaluation method with no reference does not require the original video information,directly rely on the video to be tested to obtain the video quality.It also has the advantages of flexibility,real-time,operability,wide range of applications and so on.So the study of No-reference video quality Evaluation method in this paper is of great significance.Video distortion can be caused in the video’s collection,coding,transmission and decoding process,and study has found that the main cause of video distortion is video codec process.According to the different reference information obtained before and after decoding,the No-reference video evaluation can be divided into Compressed Domain quality evaluation and Pixel Domain quality evaluation.In the compressed domain,the video is evaluated by the information obtained from the compression coding,such as quantization,motion vector and so on.In the Pixel Domain,the video is evaluated by the distortion type,such as Blockiness and Blurring,which exist in the video.The reference information of the two domains is different,and the evaluation results are different.The evaluation method in this paper combines the two-domain information and fused them to improve the accuracy of the evaluation results.Based on the research background,this paper has carried on thorough research on the way of MPEG coding.This paper studies video stream structure in the MPEG codec process and the effects of different video frames on the other frames in GOP.MPEG encoding video sequence is composed of multiple GOP,and GOP is composed of I frame,P frame and B frame.The coding of the three frames is different,and the damage of each frame has different effect on the GOP.As the bidirectional prediction frame,the damage of B frame will not affect other frames.So,the B frame is removed in the evaluation process,and I frame and P frame are reserved,so that to reduce the complexity of computation.Then,different weights assigned to I frame and P frame according to their influence.This procedure improves the accuracy of the evaluation results.According to the above two points,this paper proposes a GOP quality evaluation mechanism based on the perception of the location of damaged frame.Based on the evaluation mechanism,and combined the Blockiness and Blurring of the Pixel Domain,this paper proposes a GOP-based No-reference video quality evaluation method.Finally,in the experimental part,by comparing the proposed method and the classical algorithm,this paper proves the superiority of GOP-based No-reference video quality evaluation method.By comparing PB,GOP+PB,FISH and GOP+FISH algorithms,the feasibility of the GOP quality evaluation mechanism based on the perception of the location of damaged frame is proved.The comparison between this algorithm and the Blockiness and Blurring algorithm proves the validity of this algorithm in combining Blockiness and Blurring in the pixel domain.