| 引用本文: | 李杰,刘怡光,都双丽,徐振宇.基于层次化及最小二乘的精确图像配准[J].控制理论与应用,2017,34(6):811~819.[点击复制] |
| Li Jie,Liu Yi-guang,Du Shuang-li,Xu Zhen-yu.Precise image matching via multi-resolution analysis and least square optimization[J].Control Theory and Technology,2017,34(6):811~819.[点击复制] |
|
| 基于层次化及最小二乘的精确图像配准 |
| Precise image matching via multi-resolution analysis and least square optimization |
| 摘要点击 2054 全文点击 1819 投稿时间：2016-07-28 修订日期：2017-08-17 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2017.60560 |
| 2017,34(6):811-819 |
| 中文关键词 图像配准 大尺度变换 层次化结构 对数极坐标傅立叶变换 最小二乘 |
| 英文关键词 image matching large similarity transformation multi-resolution log-polar mapping based Fourier transform least square |
| 基金项目 国家自然科学基金 |
|
| 中文摘要 |
| 针对传统对数极坐标傅立叶变换(log-polar mapping based Fourier transform, LPMFT)在大尺度、大旋转及大平移变换情况下不能精确估计图像对之间的变换参数, 提出基于层次化及最小二乘的图像配准方法(multi-resolution analysis and least square optimization, MALSO): 首先, 使用小波变换将图像分解为多分层结构, 并将每层的低频部分作为待匹配图像; 其次, 在每层中, 引入窗口函数及自适应滤波函数以减少谱泄漏, 混叠及插值误差的影响; 最后, 构建一个代价函数, 并通过最小二乘法求解最优参数. 实验表明, 该方法既满足大尺度, 大旋转及大平移参数准确估计要求, 又比LPMFT对遮挡更具鲁棒性, 有一定的理论及应用价值. |
| 英文摘要 |
| When subjected to the large scale, large rotation and large translation, the classic log polar mapping based Fourier transform (LPMFT) suffers from misregistration. Motivated by this problem, a novel approach named multiresolution analysis and least square optimization (MALSO) is proposed: firstly, original images are decomposed into multi-resolution levels by wavelet transform, and only the low frequency part of each level is chosen as the matching candidate; secondly, to alleviate the influence caused by leakage, aliasing and interpolation error, in each level, window
function and adaptive filtering technique are introduced into LPMFT; finally, for obtaining a set of optimal parameters, a cost function is carefully established which is approximated by least square optimization method. Experimental results show that the proposed approach not only minimizes the influence of the large scale, large rotation and large translation, but also achieves a better registration performance than the classic LPMFT method for occlusion image pairs. |
|
|
|
|
|