Title: Image-based Modelling and Rendering

Abstract:
IBMR (Image-based Modelling and Rendering) interests both computer graphics and vision communities. It is probably more ambitious from computer vision's point of view, as by IBMR, it stands for 'real' image-based and 'handheld' capturing. In this seminar, will show some view interpolation, morphing, reconstruction, (generalised) mosaicing and plenoptic modelling examples. Then, I will present a working and developing system of IBMR which tackles the two most difficult problems of IBMR: pixel matching and visibility handling.

The presentation of the system is organised as follows. We first describe a new quasi-dense pixel matching algorithm based on a best-first match propagation strategy. Then we introduce the joint view triangulation (JVT)---a novel representation for collections of images that handles the visibility and occlusion problems created by the parallaxes between the images. The joint view triangulation is built from matched planar patches regularized by local smooth constraints encoded by plane homographies. We give an incremental edge-constraint construction algorithm. Finally we present a pseudo-painter's rendering algorithm for the joint view triangulation, and demonstrate the performance of these methods experimentally.

Biography:
Long QUAN is a CNRS senior research scientist with INRIA, France. He received the Ph.D. degree in computer science from INRIA, France in 1989 and became a permanent researcher since 1990. His research interests include 3D vision geometry and image-based modelling and rendering. He has intensively worked on 3D vision geometry in the past decade, and made a series of theoretical contributions. In application fields, recently with his students, he has been developing a working image-based modelling and rendering system, which is probably the first of its kind in the world. Actually he is also leading two european consortiums in image-based modelling and rendering fields.