基于薄板样条配准和全局相似的图像交互拼接  

Image interactive stitching based on thin plate spline registration and global similarity

作　　者：黄华峰 陈飞[1] 曾勋勋[1] Huang Huafeng;Chen Fei;Zeng Xunxun(College of Mathematics and Computer Science,Fuzhou University,Fuzhou,350108,China)

机构地区：[1]福州大学数学与计算机科学学院,福州350108

出　　处：《南京大学学报（自然科学版）》2021年第5期828-837,共10页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(61771141);福建省教育厅中青年教师教育科研项目(JAT190020)。

摘　　要：图像拼接旨在将多张具有重叠区域的图像拓展成视野更加宽广的全景图,当输入的图像不满足原本严格的成像前提的时候,往往无法进行拼接,即使强行拼接,也可能造成不自然的扭曲以及影响观感的伪影.以往的拼接算法通过计算刚性变换来实现对齐,而通过计算非刚性变换可以在一定程度上提高对齐效果.受医学图像使用人工标记点配准的启发,提出一种基于薄板样条配准和全局相似的图像交互拼接的方法.首先通过交互标记点计算非刚性变换将目标图像和参考图像进行初步对齐;其次,引入全局相似性缓解非重叠区域的投影失真;最后使用优化的交互图切算法进一步消除伪影,并提升整个算法的鲁棒性.实验证明,在许多商用软件无法拼接的场景下,该算法仍然可以成功拼接,并且产生更加自然清晰的结果.Image stitching aims to expand multiple images with overlapping areas into a wider panoramic view.When the input image does not meet the original strict imaging premise,it is often unable to be stitched.Even if possible,the result may cause unnatural distortion and artifacts which affect the perception.The previous stitching algorithm realizes alignment by calculating rigid transformation.In fact,the alignment effect can be improved to some extent if the non⁃rigid transformation is calculated.Inspired by use of artificial landmarks to register medical images,this paper proposes an image interactive stitching based on thin plate spline registration and global similarity.Firstly,the target image and the reference image are aligned preliminarily by calculating the non⁃rigid transformation of interactive landmarks.Secondly,global similarity is introduced to alleviate projection distortion in non⁃overlapping regions.Finally,the optimized interactive graph cutting algorithm is used to further eliminate artifacts and improve the robustness of the whole algorithm.Experiments show that this algorithm can still work successfully in many scenarios where commercial software cannot be stitched together,and produce more natural and clearer results.

关 键 词：图像拼接 图像对齐 标记点 薄板样条配准 全景图 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]