Median Gauss Laplace Blind Video Quality Prediction

Digital video quality assessment in the field of video compression, processing, and video communication plays a very important role.Video quality assessment(VQA) is on the basis of image quality assessment(IQA) gradually developed, there are a lot of advanced image quality assessment method to inspire the innovation and improvement of video quality evaluation method, but due to the video motility it is different from the image quality evaluation?method usually divided into full reference(FR), reduce reference(RR) and no reference(NR) three types of evaluation methods?Although no reference evaluation method has been gradually developing in recent years,its performance still can not match with the performance of full evaluation methods, and it still has great rooms to improve.Blind VIDEO quality prediction(VIDEO BLIINDS) is a no reference video quality evaluation algorithm in recent years with prominent performance in international?This algorithm testing on the LIVE VQA databases shows that its performance close to the top FR and RR VQA algorithms.In this paper, after analysis of the algorithm, we improve and optimize the multifaceted performance?And the new algorithm testing on the LIVE VQA databases shows a better performance.In this paper, the main results are as follows:1.In this paper,we propose a improved model of gauss Laplace filtering(LOG)processing natural scene statistical(NSS) properties of space and time for a single frame image.By analyzing the natural scene statistics(NSS)model,we found that the extracted edge information of each sub-block losing a lot of useful information.And we adopt gauss Laplace filter(LOG) instead of gaussian filtering to enhance the edge of the inside of each sub-block information extraction.Thus we get a new time and space natural scene statistics(NSS) model.By applying it to Video Bliinds,we get a new median LOG-NVS model.2.In this paper,we propose a improved model of gauss Laplace filtering(LOG) processing natural scene statistics(NVS) of space and time for videos.We adopt gauss Laplace filtering(LOG) natural video scene statistics(NVS)model for videos,and adopt median computation method to optimize the NVS eigenvalues.In turn,we makes the final video quality evaluation results more close to human subjective opinion scores.3.In this paper,we propose a new linear kernal support vector machine regression(SVR) prediction model.Under the original ksvm model,we introduced e1071 package to replace kernlab package,which making convenience for the further optimization of the model.And with no affecting on the results we simplifying the original model of a large number of repeated data,and the algorithm is added the subjective opinions about videos in test score without affecting the property of NR,making the experimental data more convenient for studying,contrasting and storage.