结合视觉舒适度的无参考立体视频稳像效果评价  

Non-Reference Assessment of Stereoscopic Video Stabilization Joint Visual Comfort

作　　者：吴剑荣 黄华[1] Wu Jianrong;Huang Hua(School of Artificial Intelligence,Beijing Normal University,Beijing 100875)

机构地区：[1]北京师范大学人工智能学院,北京100875

出　　处：《计算机辅助设计与图形学学报》2024年第9期1341-1350,共10页Journal of Computer-Aided Design & Computer Graphics

摘　　要：立体视频稳像效果评价是评价立体视频稳像算法性能的有效途径.针对当前缺乏立体视频稳像效果客观评价方法的问题,提出一种结合视觉舒适度的无参考立体视频稳像效果评价方法.将立体视频稳像前后的运动平滑度和视觉舒适度变化作为视频稳像前后的变化特征,结合主观评价训练得到立体视频稳像效果评价的支持向量回归模型;回归模型通过学习立体视频稳像前后的变化特征与主观评价结果之间的关系,最终获得直接评价任意立体视频稳像效果的能力.使用收集的55条仿真视频训练模型,并在10条真实视频上进行实验的结果表明,所提方法的稳定性较好,在视频量达到180条时,模型的评价结果趋于稳定,且模型评价结果与主观评价结果相关性达到93%,可用于立体视频稳像效果的客观评价.Assessment of stereoscopic video stabilization is an effective way to evaluate the performance of stereoscopic video stabilization algorithm.For the deficiency of stereoscopic video stabilization assessment method,a novel non-reference assessment scheme is proposed joint visual comfort.The variation of motion stability and visual comfort are employed to represent variation characteristics between stereoscopic video and the stabilized video,and a support vector regression model is trained for stereoscopic video stabilization as-sessment,combining with subjective assessment.The regression model get ability that can correctly assess stereo video stabilization algorithm’s product by learning the map between variation characteristics and subjec-tive assessment.55 simulation videos were collected to train the model.The experiments on 10 real videos demonstrate that the proposed schema has good stability,the model is convergent with the train video’s number up to 180.The result’s correlation between model and human subjective assessment is 93%,suggests the model can be used for object assessment of stereoscopic video stabilization.

关 键 词：立体视觉 视频稳像 视频质量评价 视觉舒适度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]