学术报告第201504期

Min-Cut and Max-Flow: Old and New

Nowadays the graph optimization-based theory of min-cut is widely  accepted as the most popular approach to image segmentation, which formulate and  solve the problem of segmenting images in the manner of global optimization. In  addition, a fast algorithm to min-cut can be developed based upon the framework  of max-flow which was demonstrated to be dual or equivalent to min-cut  in mathematics. Recently, the min-cut problem is reformulated and studied in the  spatially continuous setting, i.e. the continuous min-cut problem, which can be  globally solved by convex optimization and shows great advantages over the  classical graph-based min-cut/max-flow in both theory and practical results. As  the dual pair of min-cut and max-flow, we show a new dual optimization theory to  the continuous min-cut, so-called the continuous max-flow theory. For this, we  develop a series of continuous max-flow formulations which derive a set of  novel fast algorithms based on the new multiplier augmented theory. Moreover, we  show their applications to the new global-optimization-based time-implicit  level-sets and medical image analysis.

主讲人：

Name:  Jing Yuan

Research Scientist, Robarts Research Institute, Western University,  Canada

Adjunct Research Professor, Medical Biophysics Dept., Schulisch Medical  School, Western University, Canada

Dr. Jing Yuan obtained his PhD in the Dept. of Mathematics and Computer  Science, Heidelberg University, Germany. Currently, he served as the program  chairs and reviewers of the top conferences of computer vision, medical image  processing and applied mathematics: MICCAI, CVPR, EMMCVPR, ECCV, SSVM  etc. 

报告时间及地点：

2015年1月29日下午2：30  逸夫科技楼9楼报告厅