报告题目：Minimal Paths for Image Segmentation and Tubular Structure Tracking

主 讲 人：陈 达

单 位：山东省人工智能研究院

时 间：9月5日15:30

腾 讯 ID：329 234 645

摘 要：

The minimal geodesic technique was introduced about 20 years ago as an efficienttool for image analysis through the Eikonal PDE framework. As an important advantage, the minimal geodesic model is capable of finding the global minimum of the geodesic active contour energy, thus can avoid unexpected local minimization. Minimal geodesic paths have been used for long to address theproblems of image segmentation and tubular structure tracking, where the features of interest, i.e, object boundary and tubular centerlines, can be modeledas minimizing geodesic curves. The solutions to those problems can be regardedas a way to find a curve globally minimizing the geodesic active contours energy, done by solving the corresponding Eikonal PDE using the fast and efficient Fast Marching method.

In contrast to find the global minimum of a simplified active contour energy, we have recently extended the minimal geodesic model to cover all kinds of active contour energy terms. Through designing adequate geodesic metrics , we now are able to estimate minimal geodesic paths according to various active contours terms, involving curvature penalization and region-based homogeneity term. We will present the mathematical background as well as concrete applications to biomedical and natural images.

简 介：

陈达，于2017年3月获得巴黎文理研究大学应用数学博士学位, 2016年10月至2019年3月分别在巴黎多芬纳大学和巴黎国立眼科医院从事博士后研究工作，并于2019年5月入职山东省人工智能研究院, 副研究员。主要研究兴趣包括基于变分法和偏微分方程的图像分析、最小测地线模型研究及其在医学图像处理中的应用、形变模型和统计学习模型等，部分相关研究成果已经在计算机视觉和图像处理领域的主流杂志和会议上发表，如IJCV，IEEE TIP，IEEE TBME，JMIV和CVPR，BMVC等。