|
New:
|
|
Simultaneous Camera Pose and Correspondence Estimation in Cornerless Images accepted by ICCV 2009.
(with Wen-Yan Lin, Guo Dong, Loong-Fah Cheong and Chye-Hwang Yan)
[pdf]
|
|
A Projective Framework for Radiometric Image Analysis accepted for oral presentation by CVPR 2009.
(with Todd Zickler)
[pdf]
|
|
Photometric Stereo and Weather Estimation Using Internet Images accepted by CVPR 2009.
(with Li Shen)
[pdf]
|
|
Image-based Façade Modeling accepted by SIGGRAPH
ASIA 2008. (with Jianxiong Xiao, Tian Fang, Peng Zhao, Eyal Ofek, Long Quan)
[pdf]
|
|
Single Image Tree Sketching accepted by SIGGRAPH
ASIA 2008. (with Tian Fang, Jianxiong Xiao, Peng Zhao, Long Quan)
[pdf]
|
|
Intrinsic Image Decomposition with Non-Local Texture Cues
to appear at CVPR 2008. (with Li Shen and Stephen Lin)
[pdf]
|
|
Research Projects |
|
Image based modeling
|
|
Image-based modeling uses images from different viewpoints to build appearance model
of an object. Previous methods on image-based modeling are very successful at recovering
camera poses and a set of 3D points of the object from images. But these recovered 3D
points are still unstructured. We have worked on object reconstruction,
which is to organize these unstructured 3D points and build high quality texture mapped mesh model from them.
|
|
Photometric Stereo
|
|
Photometric stereo uses images from different illumination conditions to build appearance
model of an object. Conventional methods have two limitations. Firstly, surface
shape can only be recovered at the resolution of input image. Secondly, illumination
conditions should be recorded to recover surface surface correctly. We propose a
super-resolution and an auto-calibration algorithm to improve photometric stereo
from these two aspects.
|
|
Multiresolution Reflectance Modeling
|
|
The reflectance of a surface depends upon the resolution at which it is imaged.
At low resolution or sharp viewing angles, the surface area within a pixel may correspond
to that of several pixels at a higher resolution. The BRDF of a low-resolution pixel
thus is effectively a combination of the BRDFs of several higher-resolution pixels.
To address the resolution dependency
of reflectance, we propose a framework for real-time rendering of multiresolution
reflectance that is suitable for a wide range of parametric BRDFs. |
|
Reflectance Separation
|
|
When light is reflected on a surface, it can undergo two different procedures, which
lead to distinct reflectance components, i.e. diffuse and highlight. Rays of different
component carry different information about the scene. Most of existing computer
vision systems, e.g. tracking, segmentation, reconstruction, are designed for purely
diffuse surface and can easily be fooled by highlights in images. Therefore, it
is important for a computer to distinguish these components from images. Some previous
works on this separation require multiple images as input, which limits
the application. We tackle the problem with a single input image via color space
analysis. |
