随着视频重定向方法的日渐发展,对视频重定向质量客观评价算法的要求愈高.现存的视频重定向质量客观评价算法均以Scale Invariant Feature Transform flow(SIFT-flow)稠密匹配为基础.当原始帧与重定向帧尺寸不一致时,SIFT-flow稠密匹配会产生错误,客观评价算法性能下降.在匹配过程中,逆向重建网格存在未匹配区域.基于上述情况,提出一种基于逆向重建和运动轨迹偏移的视频重定向质量客观评价算法.该算法通过减小目标函数平滑项的权重和删除黑边减少了错误匹配的情况,并提出网格损失率(Grid Loss Ratio,GLR)衡量未匹配区域的空间失真,且使用网格运动轨迹偏移量减少对SIFT-flow稠密匹配的依赖.与Multimedia Lab(ML)主观数据库的肯德尔相关系数达到0.593,其标准差为0.295.与其他算法相比,该算法能更好地评估重定向过程的失真,更符合人们的主观评价结果.
Video Retargeting Quality Assessment Based on Reverse Reconstructed and Motion Trajectory Offsets
With the development of video retargeting methods,the requirements for objective video retargeting quality assessment algorithms are getting higher and higher.The existing video retargeting quality objective assessment algorithms are based on Scale Invariant Feature Transform flow(SIFT-flow)dense matching.However,when the original frames and the retargeted frames are not the same size,SIFT-flow dense matching generates errors and the performance of the objective assessment algorithm degrades.At the same time,there are unmatched regions in the reverse reconstructed grid during the matching process.Based on the above situation,a video retargeting quality objective assessment algorithm based on reverse reconstruction and motion trajectory offset is proposed.The algorithm reduces the false matches by reducing the weight of the smoothing term of the objective function and removing the black edges,and proposes a Grid Loss Ratio(GLR)to measure the spatial distortion of the unmatched regions,and uses the grid motion trajectory offset to reduce the dependence on SIFT-flow dense matching.The Kendall correlation coefficient with the Multimedia Lab(ML)subjective database reaches 0.593,and its standard deviation is 0.295.Compared with other algorithms,this algorithm can better assess the distortion of the retargeting process and is more consistent with people's subjective evaluation results.
video retargeting quality assessmentobjective assessmentreverse reconstructed gridmotion trajectory offset
万方数据
维普
