My research interests include graphics and modeling, image processing, shape estimation, stereo vision, machine learning and neural networks. My primary research focus is on the object recognition problem, specifically, identifying and locating objects from intensity images or range (depth) images. Object modeling is an implicit but extremely important aspect of recognition. In fact, the object recognition problem touches on many fundamental aspects of computer vision, graphics and object modeling. Through my research, I have acquired a broad understanding of the fundamental problems in these fields.
Computer vision is plagued by unfavorable combinatorics, uncertainty in measurements, and, in general, ill-posed problems. This leads to solutions with many implicit assumptions, heuristics, and implicit restrictions. My research philosophy is that a formal understanding of the problems and solutions is necessary to build systems that are robust and at the same time efficient. Once the problems, solutions, and algorithms are formalized, it is then possible to develop an improved or new solution using well-understood mathematical techniques. Furthermore, we can then characterize the applicability and performance of the algorithm. I am also a pragmatic researcher; I believe that analysis of the successes and failures of experimental systems is required to elucidate the mathematical nature of problems in practice.