基于几何特征相似度评价的跨尺度显微图像配准(英文)
Cross-scale image registration based on geometric feature similarity evaluation for atomic force microscope and optical microscope
摘要点击 69  全文点击 31  投稿时间:2019-12-04  修订日期:2020-03-09
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2020.90989
  2020,37(9):1913-1922
中文关键词  原子力显微镜  光学显微镜  几何特征  相似度评价  图像配准
英文关键词  atomic force microscope  optical microscope  geometric feature  similarity evaluation  image registration
基金项目  国家自然科学基金重点项目(61633012).
作者单位E-mail
武毅男 南开大学机器人与信息自动化研究所 wuyn@nankai.edu.cn 
方勇纯 南开大学机器人与信息自动化研究所 fangyc@nankai.edu.cn 
樊志 南开大学机器人与信息自动化研究所  
王超 南开大学机器人与信息自动化研究所  
刘存桓 南开大学机器人与信息自动化研究所  
胡子琦 南开大学机器人与信息自动化研究所  
中文摘要
      原子力显微镜能够在光学显微镜的协助下, 克服自身成像范围的限制获得更大的成像视野, 同时保证纳米 级的成像精度. 该方法需要实现原子力显微镜成像结果在光学视野中的精确定位, 而解决该问题的关键是进行两种 显微图像之间的准确配准. 因此, 本文提出了一种基于几何特征相似度评估的跨尺度图像配准算法, 为进一步在原 子力/光学显微镜共焦系统中实现精确定位和成像提供了基础. 具体而言, 本文首先利用原子力显微镜的探针在样 品表面进行压印, 刻画出固定尺寸的几何图案,用以标定原子力显微镜和光学显微镜成像尺度之间的比例, 为图像 配准提供先验知识. 随后, 本文设计了一种先进的图像处理算法, 分别提取原子力/光学显微镜图像中几何图案的特 征, 并将其存储为内角向量和边长向量. 最后, 基于成像尺度比例和几何特征, 提出了一种新型的几何特征相似度 评价函数, 通过对内角特征相似度和边长特征相似度进行加权融合, 实现高精度的跨尺度显微图像配准. 实验部分 针对四种不同几何图案进行图像配准, 并对实验结果进行详细分析, 验证了本文方法的良好性能.
英文摘要
      An atomic force microscope (AFM) is able to overcome the limitation of its own imaging range while ensuring high imaging accuracy with the assistance of an optical microscope (OM). However, it is difficult and necessary to realize accurate positioning of AFM images in optical field of view by linkage control between an AFM and an OM, and the key to solving this issue is to achieve AFM and OM image registration. Therefore, in this paper, a geometric feature similarity evaluation based cross-scale image registration algorithm is proposed for AFM and OM imaging, which provides the basis for further precise positioning and imaging in the AFM and OM confocal system. To be specific, to facilitate accurate image registration, the ratio between AFM and OM imaging scales is first calibrated by utilizing a known size pattern imprinted by an AFM probe, which is then applied as a priori knowledge for image registration. Furthermore, an advanced image processing algorithm is designed to extract the geometric feature for AFM and OM images, which are stored as inner angle vectors and side length vectors. Moreover, based on the calibrated scale ratio and the geometric feature, a novel similarity evaluation function is proposed to achieve cross-scale image registration with high accuracy. Experiments and analysis of different imprinted geometries are implemented to demonstrate the good performance of the proposed method.