| Compared to the traditional two-dimensional flat video,stereoscopic videos are gaining increasing amounts of acceptance and attention via providing a more shocking visual impact and realistic experiences.How to provide viewers with more realistic,more convenient and more comfortable three-dimensional visual experience is a key problem of the current three-dimensional video technology.However,due to the technical restriction of current three-dimensional video content and display technology,resulting in the visual uncomfortable physiological phenomenon during the audience watches stereoscopic video,which greatly hindered the popularization of three-dimensional video.Unfortunately,the objective standard of stereo video comfort evaluation is no uniform,and the existing objective prediction model can't effectively reflect the comfort level of stereoscopic video.Thus,it is important to establish an objective visual comfort prediction model to calculate the comfort level of stereoscopic video.For these reasons,this thesis mainly focuses on how to effectively combine the stereoscopic visual significance to establish the comfort evaluation model of the stereo image / video,and then reflect the objective comfort value of the stereoscopic image / video comfort through the model.The main contents of this paper are as follows:We propose a saliency detection model based on binocular multidimensional sensing.First of all,SLIC algorithm is used to segment the input color image for super pixel segmentation,obtain the regional segmentation map is similar with super pixels near the regional characteristics in the calculation of two-dimensional space saliency,which can effectively improve the efficiency of subsequent saliency detection.Next,we combined with human binocular vision perception characteristics in depth saliency calculation,we not only consider the absolute parallax feature in the disparity map,but also calculate the local disparity features and the depth of the contrast salient features,which is possible to improve the accuracy of the depth saliency detection.And then,we combine the visual fatigue characteristics of the human eye in the process of saliency fusion and takes care of its disparity distribution to obtain the stereoscopic image salient region distribution in accordance with the human attention mechanism.On the basis of this,a new model of stereoscopic image objective visual comfort prediction is established.This model extracts important visual perceptual features in stereoscopic images and calculate the objective comfort degree of the stereoscopic images by combining the salient regions of the stereoscopic images,which effectively improve the efficiency and accuracy of the stereoscopic comfort.Based on the visual comfort model of stereoscopic image,a stereoscopic video objective visual comfort prediction model based on stereoscopic visual saliency is proposed.In the time domain saliency detection,the local region optical flow method is used to calculate the salient motion of the neighboring frames,which can effectively reduce the algorithm complexity.In the process of saliency fusion of two dimensions,depth and time domain,BP neural network training is used to obtain the optimal weight distribution of the saliency map of each dimension and obtains the saliency region distribution model of the stereoscopic video and divides and extracts the salient target,which improve the accuracy of subsequent objective visual comfort prediction.In the process of time-domain visual perception feature extraction of stereo video,the plane motion features and the motion of disparity in depth direction are extracted respectively,which can more accurately reflect the visual comfort degree of the human eye when watching stereoscopic video.Experimental results show that the objective evaluation model proposed in this paper has a good performance,and the objective prediction result is roughly the same as the subjective comfort score,which can accurately reflect the viewing comfort of the stereoscopic video. |
PDF Full Text Request