Matting is the classic computer graphics and vision problem of
separating an object from its background by computing a "matte" image that
marks the foreground region. Since 1958 the film
industry has used special background screens to simplify matting.
We were the first to automatically compute mattes for
video of natural scenes without special backgrounds or lighting.
We also merged those techniques with
special backgrounds to generate exceptionally high quality and
real-time mattes in the studio.
Imaging is a measurement technique employed throughout the physical
sciences. Often one cannot directly measure the desired quantity
and computation is required to process the captured image.
Analysis of image registration noise due to rotationally dependent aliasing. Stone, Tao and McGuire. VCI, 2003
Techniques
for Multiresolution Image Registration in the Presence of
Occlusions. McGuire and Stone. Trans. GeoRS 2000
The Translation Sensitivity of Wavelet-Based Registration, Stone, Le Moigne and McGuire. PAMI 1999
The translation sensitivity of wavelet-based registration. Stone, Le Moigne and McGuire. AIPR 1998
Techniques for multi-resolution image registration in the presence of occlusions. McGuire and Stone. Image Registration Workshop, 1997
Shader Authoring
This project investigates new paradigms for programming graphics
hardware without writing source code. This allows artists and hobbyists
to create more compelling content, and empowers programmers to develop
applications faster.
Stylized images like cartoons can communicate shape and motion
better than photorealistic images. This makes them prefferable for
many applications like entertainment and medicine.
However, stylized images have proven more challenging to produce
than photorealistic ones.
This project investigates new algorithms for rendering 3D objects in artistic styles using graphics hardware.
We described several new optimizations for the shadow volume
method of shadow casting that then appeared in video games
like Doom 3, and characterized the properties of shadow volumes
to introduce new theoretical and practical bounds on runtime
performance.
G3D is a commercial-grade 3D Engine available as Open Source (BSD License). It is used in games, tech demos, research papers, military simulators, and university courses. It provides a set of routines and structures so common that they are needed in almost every graphics program. It makes low-level libraries like OpenGL and sockets easier to use without limiting functionality or performance. G3D gives you a rock-solid, highly optimized base from which to build your application.
Thanks to Brown University and Mitsubishi Electric Research Laboratory, where some of these projects were performed. Thanks to NVIDIA, ATI, and Autodesk for supporting this research.