Sing Bing Kang
Title: Image-Based and Sketch-Based Modeling of Plants and Trees
Plants and trees are among the most complex natural objects. Much work has been done attempting to model them, with varying degrees of success. Over the past few years, I've been involved in efforts to model plants and trees using images and sketches. In this keynote, I will describe these projects.
Image-based approaches have a distinct advantage in that the resulting model inherits the realistic shape and complexity of a real plant or tree. We use different techniques for modeling plants (with relatively large leaves) and trees (with relatively small leaves). Both techniques start with the same initial step of structure from motion. In our plant modeling system, each leaf is generated by segmenting, after which it is fit to a deformable leaf model. The branches are interactively added, guided by the pre-computed 3D points. To model trees, we instead use the shape patterns of visible branches to predict those of obscured branches. Because the tree leaves are small and many, they are generated through back-projection from the source images onto the reconstructed branches.
Sketch-based systems, on the other hand, are more intuitive to use. I will describe a sketch-based approach that converts a user's freehand sketch of a tree into a full 3D model that is both complex and realistic-looking. Our system does this by probabilistic optimization based on parameters obtained from a database of tree models. Branch interaction is modeled by a Markov random field, subject to the projection constraint. Our system then uses self-similarity to add new branches, before finally populating all branches with leaves of the user's choice.
I will show reconstructions of a variety of plants and trees to substantiate the techniques described. Finally, I will offer my thoughts on improving our systems and on the remaining challenges associated with plant and tree modeling.
Sing Bing Kang received his Ph.D. in robotics from Carnegie Mellon
University, Pittsburgh, USA in 1994. He is currently Principal
Researcher at Microsoft Corporation working on image and video
enhancement as well as image-based modeling. Sing Bing has co-edited
two books ("Panoramic Vision" and "Emerging Topics in Computer
Vision") and co-authored two books ("Image-Based Rendering"
"Image-Based Modeling of Plants and Trees"). He has served as area
chair and member of technical committee for the major computer
vision conferences. He has also served as papers committee member
for SIGGRAPH and SIGGRAPH Asia. Sing Bing was program co-chair for
ACCV07 and CVPR09, and is currently Associate
Editor-in-Chief for IEEE TPAMI and IPSJ Transactions on Computer
Vision and Applications.
Sing Bing's homepage