题 目：Nonconvexweighted variational metal artifacts removal via convergentprimal-dual algorithms
内容简介：Directreconstruction through filtered back projection engenders metalartifacts in polychromatic computed tomography images, attributed tohighly attenuating implants, which further poses great challenges forsubsequent image analysis. Inpainting the metal trace directly in theRadon domain for the extant variational method leads to strong edgediffusion and potential inherent artifacts. With normalization basedon pre-segmentation, the inpainted outcome can be notablyameliorated. However, its reconstructive fidelity is heavilycontingent on the precision of the pre-segmentation, and highlyaccurate segmentation of images with metal artifacts is non-trivialin actuality. In this paper, we propose a nonconvex weightedvariational approach for metal artifact reduction. Specifically, inlieu of employing a binary function with zeros in the metal trace, anadaptive weight function is designed in the Radon domain, with zerosin the overlapping regions of multiple disjoint metals as well asareas of highly attenuated projections, and the inverse square rootof the measured projection in other regions. A nonconvexregularization term is incorporated to further enhance edge contrast,alongside a box-constraint in the image domain. Efficient first-orderprimal-dual algorithms, proven to be globally convergent and of lowcomputational cost owing to the closed-form solution of allsubproblems, are devised to resolve such a constrained nonconvexmodel. Simulation experiments are conducted with comparisons to othervariational algorithms, validating the superiority of the presentedmethod. Especially in comparison to the reweighted JSR, our proposedalgorithm can curtail the total computational cost to at mostone-third, and for the case of inaccurate pre-segmentation, therecovery outcomes by the proposed algorithms are notably enhanced.
报告人：常慧宾
报告人简介：天津师范大学数学科学学院，研究员，主要研究领域为计算光学、医学图像处理及高性能计算。2012年博士毕业于华东师范大学，2012年-2013年在香港浸会大学从事博士后研究工作，2016年-2019年访问美国劳伦斯伯克利国家实验室。
时 间：2023年11月15日（周三）晚上20：30开始
地 点：腾讯会议：870-549-992
热烈欢迎广大师生参加!
2023年11月14日
