I am interested in automatically extracting statistical invariants from visual signals. These invariants can be physically meaningful, such as reflectance, illuminant spectra, 3D relative positions and camera parameters; and they can be transformed version of such physically meaningful invariants. The extracted, or learned information will be stored as non-parametric distributions. And they can be integrated into a Bayesian inference framework that can help an intelligent agent to further explore the world. I am working toward developing the algorithms and applying them to typical vision problems such as color constancy, object recognition and structure from motion.
In addition, I am studying texture models and its interaction with lighting effects, such as penumbra and shadows.