Statement of Research Intent

Advances in digital imaging have made it possible to put cameras almost everywhere, from cellular phones to satellites. These cameras enable us to generate large numbers of pictures of almost anything in the world. Harnessing this imaging capability in applications, such as computer graphics, monitoring, or computer vision, will require flexible models of images and the scenes that generate them. My goal is to develop machine learning techniques for building tractable models of images for applications in graphics, image processing, data mining, and image understanding. Machine learning is an exciting tool for solving computer vision problems because it provides a means for building more accurate models of images and the scenes pictured in those images. My work in recovering the intrinsic characteristics of a scene, described below, is a good example of how applying learning techniques can lead to significantly improved results on a classic vision problem.