Unsupervised Phase Unwrapping Driven by Nonconvex Optimization and Deep Image Prior
数学专题报告
报告题目(Title)：Unsupervised Phase Unwrapping Driven by Nonconvex Optimization and Deep Image Prior
报告人(Speaker)：常慧宾 （天津师范大学）
地点(Place)：后主楼1124
时间(Time)：2025年6月26日(周四) 下午15:30-16:10
邀请人(Inviter)：段玉萍
报告摘要
In the forefront of imaging technology, the limitations of modulo-2pi phase measurement significantly hinder the in-depth understanding and analysis of the underlying characteristics of images. We introduces a novel tail-minimization mechanism, which precisely identifies and corrects cumulative errors in phase estimation during iterations while intelligently excluding pixels that have already converged from subsequent computations, thereby significantly reducing computational load and accelerating the overall convergence. Further combined with the deep image prior, an unsupervised deep learning approach is given to improve the accuracy. Numerical results indicate that, particularly in handling highly complex and high-wrapping-phase data, the proposed algorithm exhibits superior recovery performance and stability.
主讲人简介
常慧宾，天津师范大学数学科学学院党委副书记、院长,天津市特聘教授（青年学者），天津青年五四奖章获得者。主要研究方向为：反问题建模与计算、图像处理、计算光学、高性能计算等，在本领域期刊如SIAM系列、IEEE汇刊等杂志以及学术会议发表学术论文五十余篇，其中2篇论文被国际晶体学联合会（IUCr）系列期刊选为封面论文，单篇论文被引最高超过200次；主持4项国家自然科学基金项目。获国际计算科学大会ICCS 2019最佳论文奖、第一届天津市数学和统计学联合年会青年学者奖等。关注计算数学与光学、医学等学科的交叉研究，提出的偏相干计算成像算法等多项工作被国际知名实验室及ScienceDaily、Phys.org等网站转载，参与开发和升级高通量成像软件，为世界多个主要同步辐射光源提供成像数据分析服务。
