Video Quality Evaluation Based On Saliency Weighting

With the continuous development of multimedia technology and the increasing number of Internet users,digital video has become an indispensable medium for people to receive information.However,while processing video information,the steps of compression,transmission and reconstruction in video data processing can cause the lack of video information,thereby distorting the video and affecting the quality of the video.Therefore,people pay more and more attention to adopting accurate and efficient quality evaluation algorithms in video processing.The evaluation method is divided into subjective and objective: since subjective evaluation comes from the intuitive perception of the human eyes,it is always regarded as the most stable evaluation method.However,it wastes much time and manpower lead to its limited application,so the study of objective methods is particularly important.Moreover,visual saliency is an important feature of the human visual system,it is often used in the field of quality evaluation.Therefore,this thesis makes some improvements and innovations to the objective method by using saliency,the main contents are as follows:(1)This thesis analyzes the significance of visual salience.Firstly,the traditional static image visual saliency model is constructed by using three simple prior knowledge of band-pass filtering,color and position.Based on this,the video motion characteristics are analyzed.The block matching motion estimation algorithm is used to improve the salient model.The visual saliency model which is more suitable for detecting the saliency of video frame sequences is introduced without sacrificing excessive computing speed.This saliency map can be used to innovate and improve the evaluation method after this article.(2)According to the frame structure characteristics of MPEG video GOP,the single frame quality score is weighted,and the GOP quality score is weighted and summed by saliency value.A video quality evaluation algorithm based on time domain GOP weighting is proposed.The performance of the algorithm is evaluated by MPEG-2 distortion type videos in the LIVE video database,and the test video verifies the superiority of the algorithm.The experimental results show that the algorithm can accurately evaluate MPEG-2 distorted video.Its performance is better than some mainstream algorithms and has better effect than directly obtaining the average of GOP quality scores.It can be seen that saliency weighting of GOP quality scoreseffectively improves the performance of the algorithm.The method better solves the problem that the importance of different frame quality scores in the human eye is different.(3)A non-reference video quality evaluation algorithm based on saliency multi-distortion features is proposed,which includes: blocking artifact,blurring and time domain entropy.At the same time,through experimental analysis,saliency values were added as new eigenvalues.In the calculation of distortion eigenvalues,Using the visual saliency model to weight the distortion eigenvalues when extracting blocking artifact,blurring effects and time domain entropy,so as to Improve predictive performance of distortion features.At last,the final quality evaluation model is constructed with several features.The test results in the LIVE video library show that the algorithm can accurately evaluate different distortion types of video,and the performance is better than some mainstream algorithms.