Home »

Research

I have conducted research into computer vision and machine learning

While a student at Cornell in 2004-2005 I implemented a method for the matching of flexible templates with cycles. The method is approximate (the problem, as I recall, can be proved NP-hard) and uses loopy belief propagation (at the suggestion of Prof Huttenlocher). This was done as part of the really, really enjoyable course CS 664 Computer Vision (link is to most recent semester).

I am interested in content based image search (see wikipedia). My own subtle contribution derives from the recognition that often what is interesting about a scene is the presence of multiples of a certain item (e.g. apples). I attempted to find these multiples by segmenting the image and clustering segments using hierarchical agglomerative clustering. Again I am indebted to Prof Huttenlocher (and Prof Felzenszwalb) for their image segmentation code, which did all the heavy lifting.