Research On No Reference Quality Assessment Of Stereo Video Based On Saliency And Sparsity

At present times,with the development of media equipment,image and video has become important source for getting information.Compared with text,images and videos are more intuitive and easy to understand.In addition,they can better convey information to people.Because of this,the technologies related to visual information processing have developed rapidly and become a research hot spot.At the same time,video can transmit more information comparing with images,which also promotes the development of video research.In these research topics,the quality assessment about stereoscopic video is closely related to human vision and has theoretical and practical significance.Therefore,it has become an important research topic.In the process of video preprocessing,compression,encoding,decoding and transmission,the video will suffer from various types of distortion inevitably,such as JPEG2000,resolution attenuation,sharpening,reduced sampling distortion and so on.Although the causes of these distortions are different from each other,they will have a bad influence on the visual perception of the human eye,so it is necessary to evaluate the distortion of the video correctly.Different from image quality assessment,videos contains both spatial information and temporal information,and temporal and spatial are related closely.Therefore,it has more obstacles about video quality evaluation.At present,mature algorithm has not been formed to solve such problem.Based on the above reasons,this paper takes video quality assessment as research topic and analyzes the specific issues of stereoscopic video quality evaluation without reference images,aiming at constructing a no reference algorithm which can be effectively evaluated and conforms to the visual characteristics of human eyes.In order to construct a scientific and reasonable evaluation system,this paper first analyzes the visual characteristics of the human eyes to the images,then implements the extraction of left and right frames at regular intervals on the stereoscopic videos,then integrates the spatial information of the left and right images through the visual sum-difference theory,combining the spatial information and the temporal information(motion information)of the videos.Next three-dimensional saliency algorithm is used to obtain spatio-temporal saliency map,and then used for the process of feature extractionDuring the feature extraction,the paper first obtains a significant dictionary from the saliency map of the video,attaining the basic atomic representation of the saliency maps based on the sparse coding algorithm,using these atoms to form the characteristics through mathematical methods.Then,the features are processed by the sparse automatic encoder,which attempts to reduce the number of the features and further extract the sparse structure of features at the same time,improving the performance of the features.Finally,the optimized features are input into the support vector machine(SVM),then fittes the subjective evaluation score of the human eye.Four evaluation indexes are used to evaluate the performance of the algorithm.In the relevant chapters of the experiment,several groups of contrastive experiments for a single variable are used to prove the rationality of the framework,and the images produced during the experiment are drawn for further principle analysis.