Blind Video Quality Assessment Based On Visual Salient Region

With the rapid development of computer technology and communication technology,various types of video services such as video conferencing,video phone have become the main way of obtaining information in people's life.However,various types of distortions are inevitably generated during the video from being generated to the receiver,which seriously affectsthe user's viewing experience.Therefore,it is of great practical significance and application value for video quality assessment methods.The non-reference video quality assessment method has become a hotspot in the field of video quality assessmentwith high flexibility,good real-time performance,low cost,and no need for any original video information.To accurately evaluate the quality of video,a blind video quality assessment method based on visual salient region is proposed and the method proposed is improved in this paper.The main results of this paper are as follows:1.Considering the impact of human visual characteristics on video quality evaluation,the visual salient region is introduced into quality assessment methods.Visual salient regionis extracted before the evaluation,removing the interference from background region.Thus the region which the human eyes concerns is evaluated.The experimental results under the LIVE video library show that the objective evaluation score after visual salient region being extracted is more consistent with human subjective perception.2.Combining the visual salient region and the spatio-temporal statistical characteristics of the video,a blind video quality assessment method based on visual salient region is proposed.By generalized Gaussian model fitting,energy distribution calculation,frequency dispersion degree calculation and basic spectrum calculation on the three-dimension discrete cosine transform(3D-DCT)coefficients of visual salient region,five spatio-temporal statistics features are extracted.The experimental results under the LIVE video library show that the extracted features can effectively characterize the video quality,and the proposed method can accurately evaluate the video quality.Compared with the existing methods,the predicted score of the proposed method is closer to the subjective score.3.In order to evaluate the video quality more accurately,the proposed method is improved.Based on the characteristics of attention of human eyes,video keyframes that are more concerned by human eyes are extracted in this paper by detecting changes in video scenes,and then texture features of key frames are extracted in the DCT domain.This feature is added as a static feature of video to the proposed method,improving the prediction performance of the evaluation system.4.Since the distortion of the video in the current mainstream video library is caused by a single cause,the distortion type of the key frame is used as the judgment basis of the video distortion type.The video distortion classification was introduced into the proposed method and tested under the LIVE video library.Experimental results show that,compared with the original method,the improved method bystatic feature and the video distortion classification improves the prediction monotonicity of 2.77% and the prediction consistency of 1.66%.Compared with other methods,the superiority of the improved method is proved.