On Variational Models for Image Registration Problems
报告题目(Title)：On Variational Models for Image Registration Problems
报告人(Speaker)：Prof. Ke Chen (Centre for Mathematical Imaging Techniques, The University of Liverpool, United Kingdom)
地点(Place)：Zoom 会议号: 660 261 88134 会议密码: 911569
时间(Time)：2020年10月28日（星期一） 16:00-17:00
邀请人(Inviter)：刘 君
报告摘要
Variational models have been used for image co-registration successfully in recent years. They offer advantages over parametric registration models when deformation is large or of unknown type. Owning to the nonconvex nature of the data fidelity term and consequently of the overall model, the coupling parameter which cannot be yet chosen in a systematic way plays a major role in determining the accuracy of registration and the correctness of the underlying map. The issue is more acute than for other imaging models (e.g. restoration or segmentation) because there exists a delicate balance between small registration errors (accuracy) and good quality of the underlying maps (correctness). We are particularly concerned with the likely scenario where some choices of the parameter may generate plausible maps and visually pleasing results but yet the registrration results are incorrect. In this talk we discuss methods to deliver diffeomorphic maps which give rise to correct registration results. Both 2D and 3D models are considered. We introduce our latest works on imposing a robustness constraint term that ensures a diffeomorphic map. While in 2D the constraint by Beltrami coefficient is well defined, it cannot be generalised to 3D directly because (non-existent 3D) complex analysis must be used. Our new idea for 3D is based on an algebraic construction that mimicks the way Beltrami coefficient characterises quasi-conformal maps in 2D. The obtained model outperforms the start of art models. Numerical results are given. Joint work with Daoping Zhang (CUHK) and Anis Theljani (Oxford).
主讲人简介
陈柯(Chen Ke)教授，博士生导师，担任如下职务：
利物浦大学数字图像技术研究中心主任(Director of Centre for Mathematical Imaging Techniques)，
英国国家工程和自然科学研究委员会利物浦医疗健康数学中心主任(Director of EPSRC Liverpool Centre for Mathematics in Healthcare).
皇家御批数学家（Chartered Mathematician）.
英国利物浦大学物理科学院副院长（国际化）。
南昌大学特聘讲座教授(2016-19)。
曾任英国利物浦大学(University of Liverpool, UK) 数学科学系主任
和大连理工大学海天学者(2006-2013)。辽宁省百人计划入选者。
杂志: Numerical Algorithms、Computational and Applied Maths、 International Journal of Computer Mathematics、Journal of Mathematical Research with Applications编委。
学习经历：早年毕业于大连理工大学，在英国曼彻斯特大学和普利茅斯大学分别拿到硕士和博士学位，以及英国里丁大学博士后。
研究领域：数值线性代数、微分方程的数值解法，数值优化和快速计算、图像模型与数值分析。先后主持英国12个研究项目，目前负责2项（1.3百万2013-2017和2.5百万英镑2015-2021）英国大型国家项目。发表杂志论文130篇，出版专著1部。
个人网址：http://www.liv.ac.uk/~cmchenke
