一种基于SURF特征的AFM图像自动拼接方法  

A SURF-BASED AUTOMATIC AFM IMAGE STITCHING METHOD

作　　者：刘笑含[1,2] 方勇纯[1,2] 任逍 张雪波[1,2] 

机构地区：[1]南开大学机器人与信息自动化研究所,天津300071 [2]天津市智能机器人技术重点实验室,天津300071

出　　处：《系统科学与数学》2014年第12期1465-1474,共10页Journal of Systems Science and Mathematical Sciences

基　　金：国家自然科学基金(61127006;61203333);天津市应用基础与前沿技术研究计划(13JCQNJC03200)资助课题

摘　　要：AFM(Atomic Force Microscope)在生命科学等领域的应用日益广泛,然而,现有的AFM由于扫描范围较小,难以得到尺寸较大的样品的全貌,针对这个问题,文章考虑AFM连续扫描时图像间存在的平移,旋转,尺度变换等关系,提出了一种基于SURF特征的AFM图像自动拼接方法,可快速准确地获取尺寸较大的样品的全貌,并保留其细节信息,以满足生命科学等领域的实际要求,具体而言,文章分析了AFM在进行连续扫描时不同图像间重叠区域的特点,提取得到了图像的SURF(Speeded Up Robust Features)特征,并设计了一种精度较高的图像双向配准方法,通过综合考虑正反两个方向的变换矩阵来得到更为准确的变换参数.在此基础上,采用渐入渐出的加权平均融合方法来消除拼接缝等问题.论文对这种AFM图像自动拼接方法进行了大量实验测试,结果表明:这种AFM图像自动拼接方法具有方便快速,准确度高等优点,可应用于生物细胞扫描来展现其全貌.Atomic force microscopes（AFMs） are applied more and more widely in the fields of life science,etc.One deficiency of current AFMs is that the scanning range is very small,which results in that the overall topography of relative large samples cannot be obtained at the nanoscale.In order to address this issue,the paper analyzes the translation,rotation and scale of the continuously scanned images,and proposes an automatic SURF-based（Speed Up Robust Features） AFM image stitching method,which yields the nanoscale overall topography of relative large samples accurately and efficiently to successfully satisfy the requirements in the areas of life science,etc.Specifically,when the AFM continuously scans different areas,there are generally some overlap among the neighborhood scanning images.For the overlapped image parts,the SURF algorithm are applied to extract the image feature pairs.Using these features,a bidirectional image matching method with high precision is proposed by taking into account the transformation matrices in two directions.Moreover,an averaging method with smoothly-changing weights is proposed to eliminate the appearance of seam lines during the image stitching.Lots of experimental results show that the proposed image stitching method is easy to use,efficient,and high accurate,which successfully obtains the overall topography of cells.

关 键 词：原子力显微镜 SURF特征 RANSAC 变换矩阵. 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]