|
Overview of active research at the RPI Vision Group. See also complete list of
publications, and download section.
Geometric Modeling and Algorithms |
Dynamic scenes and change detection (Stewart and Freedman) - Building a model
of a scene from a series of 3D scans taken from different viewpoints and at different times ... |
Manifold learning (Freedman) - Given unorganized points drawn from an unknown manifold, can you reconstruct the manifold? This problem ... |
Registration |
2D images (Stewart) - Our goal is to develop a registration algorithm capable of
aligning a wide variety of images
including image pairs taken under
substantially-different illumination ... |
Range scans (Stewart) - Many range scanners provide both sets of points in 3D and
RGB images
We are developing new algorithms to exploit this
information during 3D rigid registration ... |
Medical deformable (Stewart) - We are working toward extending the Dual-Bootstrap
to deformable registration, especially for longitudinal change detection ... |
Retinal Image Analysis |
Registration (Stewart) - The Dual-Bootstrap was original developed for retinal image registration. It has been applied to thousands of image pairs ... |
Mosaic construction (Stewart) - We have developed algorithms that build seamless mosaics of high-resolution retinal images algorithms that compute ... |
Feature extraction (Stewart) - Our feature extraction algorithms "trace" out the vascular structure of the retina
The earliest algorithm, developed ... |
Other (Stewart) - The retina project involves a wide variety of other results, including high-speed algorithms,
real-time systems, and
automatic change detection. ... |
Robust Estimation |
Robust estimation (Stewart) - In the mid- to late-1990's we did a
significant amount of work on robust
estimation in computer vision, including both the underlying theory and
the design ... |
Tracking |
Illumination-invariant tracking (Freedman) - Tracking through sharp illumination changes is a challenging task,
particularly
if the algorithm relies on photometric information. ... |
Distribution tracking (Freedman) - Can you find an object just based on colour or texture
? It seems that humans
do this all the time, and simple, task-specific algorithms do too ... |
Branch and bound techniques (Freedman)- Contour tracking can be made robust by posing
it as a mixed continuous-combinatorial optimization
problem. The combinatorial part ... |
Segmentation |
Atlas Morphping (Freedman) - Given a collection of objects along with their geometric description
(an "atlas"), we would like to find a corresponding collection in an image ... |
Model Based Segmentation (Freedman) - Given a volumetric image and a model of the object to
be segmented, segment the desired
object efficiently. New well-posed algorithms will allow ... |
Graph Cuts |
Multiscale terms (Freedman) - Multiscale terms are often seen to improve the performance of algorithms. In this work, we show how to incorporate
multiscale terms ... |
k-Pixel interactions (Freedman) - Graph cuts have proven to be extremely useful in computer vision
, as they allow one to find
the global optimum, or a strong local optimum ... |
|
|