Place：Room 402, State Key Laboratory of CAD & CG, Library and Information Center Building B, Zhejiang University Zijin’gang Campus
Title of the report ：New Formulations for Image Segmentation
Reporter：：Dr. Junyan Wang
Abstract：Image segmentation is a classic topic in computer vision and image analysis. Recently, various new models have been proposed for this task. The models include functional optimization based models, mathematical programming based models and graph-theoretical models etc. In this presentation, I will present three of my recent research works on the new formulations for image segmentation. One of these works is a new functional optimization formulation for texture segmentation, which integrates descriptor learning and Mumford-Shah model into one unified optimization model. This work is mainly focused on the the theoretical rationale of the proposed formulation. I will also introduce a new compact linear programming (LP) formulation for object segmentation. This new formulation requires significantly less computational cost with higher scalability, while achieving a similar segmentation accuracy, compared to both the common LP model and the discrete counterpart. Lastly, I will introduce one new graph cut formulation for segmentation in which edge detection is integrated. This work provides a canonical mathematical relation between edge detection and segmentation. I will also show how the model on pixels can be extended to superpixels. The resultant methods outperforms the competitive methods in accuracy with significant less computational costs.
Bio：Dr. Junyan Wang received the Ph.D. degree from the School of Electrical and Electronic Engineering in Nanyang Technological University, Singapore, in 2012. Before this, he received the B.Eng. and the B.Sc. dual degree from the School of Communication and Signal Processing and the School of Mathematics and Applied Mathematics, University of Electronic Science and Technology of China, in 2007. Currently, he is a postdoctoral research fellow at Singapore University of Technology and Design (SUTD). His research interests include interactive and automatic object segmentation, image registration, image and video analysis and applied mathematics.