Research On Video Quality Assessment Metric Based On Human Visual System

In recent years,with the rapid development of science and technology,the traditional display technology has been unable to meet people's visual needs.Especially with the popularity of Internet technology,multimedia technology has been widely used in real life.Such as virtual reality technology,video conference,video monitoring,etc.However,in the process of collecting,storing,encoding,transmitting and displaying of the video,due to the effect of video processing and communication system,the video will inevitably introduce noise or interference to reduce the video perceived quality.And the drop of the quality of video can reflect the performance of video processing system and the channel transmission quality of service.Therefore,to evaluate the quality of these videos has important practical significance and application value.In this paper,by studying the human visual system visual perception,objective quality evaluation method is proposed for stereoscopic image,monocular video and stereoscopic video.(1)A blind image quality assessment method for stereoscopic images is proposed based on histograms of oriented gradient features of cyclopean map.Considering the binocular perception of stereoscopic images of human eyes,firstly,left and right images are fused to construct a cyclopean map.Then,histograms of oriented gradient features are extracted from the cyclopean map to act as binocular features.The subjective quality image database is divided into two parts,one part is the training image and the other part is the test image.The feature of the training image is used to extracted visual dictionary based on the generalized Gaussian distribution by fisher coding.The features of test image are further encoded with the trained visual dictionary.Finally,a regression model between features and subjective scores is established via support vector regression.Experimental results show that the proposed method is effective for both symmetrical and asymmetrical stereoscopic image databases,and can achieve high consistent alignment with subjective assessment.(2)Considering the multi-channel characteristics of human visual impact on the quality of video and the motion information of video,this paper proposes a no reference monocular video quality evaluation method.To evaluate the spatial quality of video,the frame of video be decomposed to several frequency bands using the Difference-of–Gaussian(DOG)model,then extract joint generalized local binary patter feature from image bands.While evaluate the temporal quality of video,the video sequence can be divided in to spatiotemporal slices(STS)images,then the STS image also using DOG model to imitate human visual system,extract joint generalizedlocal binary patter feature from STS frequency bands.The spatial features and temporal features are fused with random forest regression model.The random forest regression model is used for mining characteristic expression.Experimental evaluations on two benchmark video quality databases demonstrate that the proposed method have achieved good performance.(3)In this paper,we propose a stereoscopic video quality assessment method which focuses on the structure information extracted from adjacent frames of single view.Spatio-temporal structural information is sensitive to both spatial distortions and temporal distortions.For two views of stereoscopic video,we firstly calculate spatio-temporal structure based local quality according to spatio-temporal gradient characteristic and chrominance information,and these local quality scores are integrated to obtain frame level scores for single view.Then,we use energy ratio map of two views as weight values to fuse the two view scores into single frame score.Finally,all the frame level scores are combined via asymmetric tracking effect.Experiments on NAMA3DS1-COSPAD1 database demonstrate that the proposed method achieves highly competitive prediction accuracy and delivers very low computational complexity.