采用改进高斯-牛顿法的视频弹性运动估计

发布时间：2018-01-22 04:39

本文关键词： 视频编码运动估计弹性模型高斯-牛顿法　出处：《软件学报》2016年11期 　论文类型：期刊论文

【摘要】：运动估计是去除视频时间维冗余的编码技术,而目前通用的平移运动模型无法有效地表示物体的局部非刚性复杂运动.为此,提出一种基于改进高斯-牛顿法的弹性运动估计方法.首先,通过分析初始迭代点对高斯-牛顿迭代结果的影响,采用基于2bit深度像素的均匀搜索预测初始迭代点;其次,通过理论和实验分析发现,不同的迭代步长对弹性运动估计/补偿性能有明显的影响,采用离散余弦变换的低频能量比率估计步长的上限,再利用黄金分割法对步长进行求精.实验结果表明,对于具有不同场景特点的视频序列,该算法始终能够保持较高的估计精度,运动补偿的平均峰值信噪比,比基于块平移模型的全搜索算法和传统弹性运动估计算法分别提高1.73d B和1.42d B.并且,该算法具有更快的收敛速度,一般仅需1~3次迭代就能取得高于传统弹性运动估计和块平移全搜索的峰值信噪比.
[Abstract]:Motion estimation is a coding technique to remove the redundancy of video time dimension, but the current translational motion model can not effectively represent the local non-rigid complex motion of the object. A method of elastic motion estimation based on improved Gao Si Newton method is proposed. Firstly, the influence of initial iteration points on the result of Gao Si Newton iteration is analyzed. The initial iteration point is predicted by uniform search based on 2bit depth pixels. Secondly, through theoretical and experimental analysis, it is found that different iterative steps have a significant effect on the performance of elastic motion estimation / compensation, and the upper limit of the step size is estimated by using the low frequency energy ratio of discrete cosine transform. The experimental results show that the algorithm can always maintain high estimation accuracy and average peak signal to noise ratio (PSNR) of motion compensation for video sequences with different scene characteristics. Compared with the full search algorithm based on the block translation model and the traditional elastic motion estimation algorithm, the proposed algorithm is 1.73 dB and 1.42 dB, respectively. Moreover, the proposed algorithm has a faster convergence rate. Generally, it takes only one or three iterations to obtain a higher PSNR than the traditional elastic motion estimation and block translation full search.
【作者单位】：辽宁师范大学计算机与信息技术学院;大连理工大学信息与通信工程学院;
【基金】：国家自然科学基金(61402214,41271422) 高等学校博士学科点专项科研基金(20132136110002) 辽宁省教育厅科学研究一般项目(L2013406) 大连市科学技术基金(2013J21DW027)~~
【分类号】：TN919.81
【正文快照】： Elastic Motion Estimation of Video Using Improved Gauss-Newton MethodSONG Chuan-Ming1,ZHAO Chang-Wei1,LIU Dan1,2,WANG Xiang-Hai11(School of Computer and Information Technology,Liaoning Normal University,Dalian 116029,China)2(School of Information and Com

【相似文献】