Searching is one of the oldest topics in computer science. It's among the most important uses of computing systems today. Recent figures suggest that up to 95% of the world's computing resource use is performing search in one form or another.
Computational searching is a huge commercial success, as evidenced by enterprises such as Google, Microsoft and Yahoo. Almost all commercial search is based on a straightforward vector space model and a mathematically simple comparison of texts according to their term frequency.
We're looking into unsolved problems, particularly metric space models and distance-based searching, where no coordinate system is available and all that can be found is the similarity of any two objects to each other.
Images and other multimedia objects have various similarities that don't translate into a conveniently indexed space. Rather, all that can be determined is how similar two images are to each other.
The unsolved question we want to be able to answer is: "Given a very large collection of images, can we efficiently find all those that are very similar to a new image?".
The Similarity and Metric Search Systems research group is led by Professor Richard Connor.