Video Temporal Slice And Its Application In Video Quality Assessment

With the improvement of the information transmission in rate,and the obvious growth about the terminal display technology,video and image replaces the text information,becomes the major media on daily communication,for its much more suitable and wealth of information.While different from other communication media,as the final receiver of the video and image,it is extremely complicated in the way that the human visual system(HVS)processes the visible information.So,it makes a great sense that qualitatively and quantitatively evaluates the distortion of the video and image to achieve the high-quality media transmission service.However,the current algorithm only provides the method to analysis the difference between the tested videos and the original ones and ignores the perceptual characteristics and strategies for HVS to process different types of distortion,also there is still much room for the improvement over the performance of the algorithm.In this paper,we propose a pooling method for distortion information,based on the recorded data from eye tracking experiment,that can be applied on the current distortion measurement to improve its accuracy in prediction and much closer to the subjective perception of human beings.For about the specific design of the algorithm,we divide the pooling model into two aspects: spatial pooling and temporal pooling,and the corresponding pooling models have been made for both to solve the problem that it has not been well considered that the eye movement has a great influence upon the result from the quality assessment.The global motion estimation has been used to predict the method of the video movement in frames and complete the segmentation and the extraction of the foreground area for each frame.Besides this,according to the result with the prediction of the movement,we first design what we believe to be a new spatial pooling model that may accurately reflect the differences in the focus of attention for the human beings viewing the different motion scene.Compared with the current global motion estimation methods,we have made several improvements in the following points: first,via the pre-analysis of the change in content between two adjacent frames,it can be achieved that automatically and adaptively select the suitable global motion estimation parameters,which can significantly reduce the computational complexity to the algorithm,without the loss in performance;second,we take the super-pixels as the basic calculation unit and replace the isolated pixels,which it has shown that the edge information of the foreground can be well preserved.Then,the global motion estimation method that we proposed can be used to solve the problems of the segmentation and the extraction of the foreground with the complicated frame movement,also the spatial pooling method based upon this method can improve the current video quality assessment algorithms in the accuracy of the prediction.In the aspect of temporal pooing,we propose three methods for analyzing,and processing recorded data from eye tracking experiment,and respectively predict the eye movement when the tester observes the tested video in three ways: the correlation between frames,the statics of the result about the clustering of each frame and the division of eye-movement speed interval.And the corresponding video temporal slice methods are proposed based upon the analysis of the above results,which is used as the data of the source in building the temporal pooling model,that can be applied to the current distortion measurement methods.The evaluation results in the relevant video quality assessment database show that the temporal pooling model based on the correlation between frames has a significant improvement,compared with the original error measurement methods.In addition,a video temporal slice algorithm based on the statics of the motion characteristics of the foreground regions of each frame can be done without the recorded data about eye movement,it can be used as a reference for other related algorithms even if the effectiveness of the method has not been approved yet.To summary up,in this paper,we propose a new method that directly evaluates the human eye movement while they observe the tested videos,which is based on the recorded data from the eye tracking experiment.We proposed three temporal slice methods by analyzing the eye tracking data in different ways,to test and verify the presume that the eye movement has indeed a great influence in spatial and temporal pooling on video quality assessment.Also,we apply the proposed pooling models in traditional error measurement methods,and it is shown that it has a great improvement on the performance of the algorithms.