东篱科研大数据发现系统（DRDS）

位置：成果数据库 > 期刊 > 期刊详情页

基于场景模式的立体图像舒适度客观评价模型

期刊名称：电子与信息学报
时间：2016.2
页码：294-302
分类：TN919.81[电子电信—通信与信息系统;电子电信—信息与通信工程]
作者机构：[1]宁波大学信息科学与工程学院,宁波315211, [2]宁波工程学院电子与信息工程学院,宁波315211, [3]南京大学计算机软件新技术国家重点实验室,南京210093
相关基金：国家自然科学基金（U1301257;61171163;61271270;61271021;61311140262）; 宁波市自然科学基金（2013A610113）
相关项目：高效3D视频感知编码及码率控制研究

关键词：立体图像, 舒适度评价, 场景模式, 双目视觉, Stereo image, Visual comfort assessment, Scene mode, Binocular vision

中文摘要：

为了预测双目立体图像内容对视觉健康可能产生的危害,该文提出一种基于场景模式的立体图像舒适度客观评价模型。根据场景中前景目标和后景区域相对于显示屏幕的凹凸性以及是否处于舒适观看区,将自然场景抽象为多种场景模式。在模式选择阶段,从视差图中自适应分割出前景目标和后景区域,根据前、后景的视差角特征确定场景所属的模式;在建模阶段,采用前、后景的视差角特征结合前景的宽度角和曲折度特征对各个场景模式分别进行建模,并量化了前、后景视差因素对视觉舒适度的影响。在IVY数据库上的实验结果表明,所提出的模型与主观感知存在较好的一致性,Pearson相关系数高于0.91,Spearman相关系数高于0.90,Kendall相关系数高于0.74,平均绝对值误差低于0.24,均方根误差低于0.32,与现有的方法相比,该文所提出的模型的评价效果更好,更接近于主观测试结果。

英文摘要：

To predict the effects induced by stereo image content on visual health, a new objective Visual Comfort Assessment（VCA） method of stereo image is proposed based on scene modes. Natural scene is abstracted as multiple scene modes according to two position states of foreground object and background region. One is the convex-concave to screen, and the other is the whether locate on zone of comfortable viewing. In the process of mode selection, disparity map is utilized to segment scene into foreground object and background region adaptively. Then, the scene＇s mode can be determined by disparity angle features of both foreground object and background region. In the modeling stage, disparity angle features of foreground object and background region, width angle and sinuosity features of foreground object are utilized to build objective VCA models in various scene modes. The experimental results tested on IVY database show that high consistency exists between the proposed model and subjective perception that Pearson linear correlation coefficient is higher than 0.91, Spearman rank-order correlation coefficient is higher than 0.90, Kendall rank-order correlation coefficient is higher than 0.74, Mean Absolute Error（MAE） is lower than 0.24 and Root Mean Squared Error（RMSE） is lower than 0.32. Compared with other existing methods, the proposed model has the better assessment performance and is much closer to the subjective assessment scores.

同期刊论文项目