Image Segmentation via Variational Model based Tailored UNet: A Deep Variational Framework

主讲人：黄忠亿 清华大学教授

时间：2025年11月21日14：30

地点：三号楼301室

举办单位：数理学院

主讲人介绍：黄忠亿，清华大学数学科学系长聘教授、博士生导师，从事计算数学与科学工程计算方面的研究。2020年获国家杰出青年基金资助，2013年获优秀青年基金资助。在多尺度数学物理问题的建模、分析和数值模拟等方面取得了一系列重要创新性成果，并成功应用于材料科学、流体力学、图像处理、金融数学、人工智能、信息论等领域。在 Mathematics of Computation, Numerische Mathematik, SIAM 系列杂志, Journal of Computational Physics 等国际顶尖杂志和IEEE等国际会议上发表论文百余篇，受到国际同行好评。

内容介绍：Traditional image segmentation methods, such as variational models based on PDEs, offer strong mathematical interpretability and precise boundary modeling, but often suffer from sensitivity to parameter settings and high computational costs. In contrast, deep learning models such as UNet—which is relatively lightweight in parameters—excel in automatic feature extraction but lack theoretical interpretability and require extensive labeled data. To harness the complementary strengths of both paradigms, we propose Variational Model based Tailored UNet (VM_TUNet), a novel hybrid framework that integrates the phase field model with the deep learning backbone of UNet which combines the interpretability and edge-preserving properties of variational methods with the adaptive feature learning of neural networks. Experimental results on benchmark datasets demonstrate that VM_TUNet achieves superior segmentation accuracy and dice score compared to traditional deep learning methods, particularly in challenging scenarios which requires fine boundary delineation.