Learning-based techniques have seen more and more successful 
applications in computer vision.  "Learning for vision" is viewed as the 
next challenging frontier for computer vision. My research has focused 
on using learning-based methods to solve computer vision problems.

Based on my experience, I characterize "learning for vision" into three 
broad categories:

(1) Learning with prior knowledge.
Using prior knowledge can significantly improve the performance of 
low-level image processing.  For example, I will present a new method 
for localizing the iris in images of human eyes based on prior knowledge 
that the iris is circular and iris texture is different from the sclera 
and pupil. This new method improves the iris localization rate 
significantly over traditional methods where only iris shape was used.

(2) Statistical inference and machine learning.
This includes generative and discriminative methods. Results will be 
presented on using new discriminative learning methods for face 
recognition and face expression recognition.

(3) Biologically-inspired methods.
Observations of characteristics of biological vision systems are important
for designing computer vision algorithms. For example, Gabor filters, 
which characterize the brain^%G�%@ receptive field, are used widely in 
computer vision. One of my research goals is to use psychophysical 
studies of human object recognition to develop better computational 
recognition methods. A new multi-perspective representation of faces 
called face cyclographs is introduced for face recognition, motivated by 
psychophysical studies that show that humans actively exploit temporal 
information for recognition. Preliminary results demonstrate that face 
cyclographs can encode rotating faces very well.