报告时间：2015年4月28日星期二上午10：30

报告地点：浙江大学紫金港校区图书信息中心B楼CAD&CG国家重点实验室402室

报告题目：Fast Discontinuity-Preserving Image and Motion Coherence:  Models and Applications

报告人：Dr.Jiangbo Lu

主持人：周昆教授

Abstract:

Resulting from light measurements of a real scene, a natural image is not a collection of random numbers simply filling up a 2D matrix. Instead, there is a rather rich amount of redundancy, self-similarity or coherence that exists locally and globally. In the same vein, visual correspondence fields or feature matches, which associate pixels (or feature points) in one image with their corresponding pixels (or feature points) in another image, possess a similar natural coherence property. However, this is just one side of the coin; on the other side, there always exist edges, boundaries or discontinuities due to e.g. the colorful yet non-flat world, independent motions of objects in the scene, and parallax induced by camera movements. As such, discontinuities in different visual “signals” are clearly roadblocks that algorithms have to effectively deal with when exploiting the coherence or smoothness property. Motivated by this, the talk is set centrally on “coherence” and “discontinuities” for images and motions. We will introduce our recent work along this line, ranging from modeling and efficient solutions to wide applications.

More specifically, the talk will start with a quick introduction of our efficient edge-aware image smoothing filters (both locally modeled and globally optimized versions) as well as their wide and concrete applications in image processing, computer vision and computer graphics. After this, the talk will move on to the key part of the journey -- dealing with two (or more) images and their related “motion” estimation, which is a fundamental problem in computer vision with many applications such as autonomous driving, image processing, computational photography, 3D scene reconstruction, place recognition and location based services, video odometry, video surveillances, and scene understanding. Strongly motivated by “an oft-told story when a student asked Prof. Takeo Kanade about the three most important problems in computer vision” [Aubry et al. CVPR’14], the talk focus here is placed on “correspondence (or alignment), correspondence, and correspondence”.  In particular, I will introduce two recent threads of work developed from our group: 1) PatchMatch filter-based dense correspondence field estimation, and 2) Bilateral function-based global motion modeling. They address the correspondence problem from different perspectives with distinctive but complementary approaches (e.g. dense field vs. scattered point sets, and discrete search vs. continuous modeling), while both share the central goal of efficiently obtaining a large number of reliable matches between a given pair of images under various challenging conditions. These difficult conditions include, for instance, matching image pairs in presence of significant geometric and photometric transformation (e.g. scale, rotation, wide baseline, large and non-rigid motions, illumination changes, image quality), across different scene contents, or containing a significant number of outliers. Again, I will give concrete examples to quantitatively and qualitatively demonstrate the advantages of our proposed approaches over several well-known methods, which include e.g. the dense SIFT flow algorithm, non-rigid dense correspondences (NRDC), and recent (piecewise constant) RANSAC techniques. Finally, the talk will be concluded by suggesting some future directions and applications.

This talk is a miniature of our two recent tutorials given or to be given at IEEE ICIP’13 and ICME’15.

Bio:

Jiangbo Lu is currently a Senior Research Scientist with the Advanced Digital Sciences Center (ADSC), a Singapore-based research center of University of Illinois at Urbana-Champaign. He also holds a joint appointment with the Coordinated Science Laboratory (CSL) of the University of Illinois. As one of the first technical staffs joining ADSC, he has been leading a research and development team in ADSC consisting of several PhD-level researchers, research engineers, application engineers and intern students to work on several funded research projects that span across basic research, applied research, as well as commercialization of technology. He has served and is serving as the PI and Co-PI for several research or technology commercialization projects, which total a funding amount of over 2.3M Singapore Dollars (SGD). Some of his research work jointly with his colleagues and project students has led to a few Best Paper Awards (or nominations), as well as ICT awards, such as the AIT Best Paper Award in the IEEE ICCV 2009 Workshop on Embedded Computer Vision. He is an Associate Editor for IEEE Transactions on Circuits and Systems for Video Technology (T-CSVT).  In this capacity, he received the 2012 TCSVT Best Associate Editor (BAE) Award. His research interests include computer vision, visual computing, image processing, video communication, interactive multimedia applications and systems, and efficient algorithms for various architectures.

Recently, three of his technologies have been licensed to a few Singaporean companies. In 2012, the live video cutout and stylization project demoed on-stage at the inaugural DEMO Asia 2012 conference won him and his team a prestigious DEMO Guru award. Two of his recent projects (Magic Touchand Social AR) have been covered by the largest newspapers based in Singapore. He holds a few US patents, international patents, technology disclosures and trade secrets as the single inventor, and contributes to a few more. Before joining ADSC in September 2009, he had earlier experience with IMEC (Leuven), Microsoft Research Asia (Beijing), and VIA-S3 Graphics (Shanghai). He received the B.S. (with honors) and M.S. degrees in electrical engineering from Zhejiang University, China, in 2000 and 2003, respectively, and the Ph.D. degree in electrical engineering from Katholieke Universiteit Leuven (KULeuven), Belgium, in 2009.