Center for Image and Vision Science
The Center for Image and Vision Science was established in 2002, affiliated with the departments of
Statistics, Computer Science, and Psychology at UCLA. Our research interest is to pursue a general unified
computational theory underlying visual perception and learning, and to build highly intelligent computer systems
which understand real world imagery and interact with people and the real environment.
Our projects span four directions:
1. Pursuing a unified theory for visual learning and modeling.
This includes learning a hierarchy of visual descriptions (vocabulary), and integrating descriptive and
generative models for visual patterns such as textures, textons, shapes, motion, face, text, clothes,
hair, plants, lighting etc over scales.
2. Pursuing a unified theory for visual inference.
This integrates techniques from statistics, computer science, and math to form general algorithms
to search for globally optimal solutions in complex and high dimensional search spaces. The key
emphasis is the unification of (bottom-up) discriminative inference with (top-down) generative
inference (MCMC and variational).
3. Pursuing new theoretical schemes for model complexity and computational complexity.
This provides metrics to measure complexity of a vision problem and here we should emphasize
ensemble complexity --- complexity averaged over the ensemble of images in contrast to worst case
complexity. The latter is almost aways NP-hard and has little relevance to real vision problems.
4. Applications: image and video parsing, 3D scene reconstruction from 2D images, graphics rendering,
visual arts, medical images, assisting visually blind and impaired population.
We acknowledge the support from NSF, NIH, ONR, ARO, NASA, Keck Foundation, Sloan Foundation,
Microsoft, Kodak, and Siemens.