Image search has a large variety of potential applications. The obvious one is of course searching images given a text query. I discuss research into how this may be done using statistical models (relevance models) to automatically annotate and retrieve images. The same techniques may also be applied to retrieve scanned historical handwritten documents and I will show an example using George Washington's documents. I will then discuss how new techniques for image search may be applied to searching printed documents in languages where good optical character recognizers do not exist. Finally, I will briefly discuss a commercial mobile image search application on the iphone which allows one to take a picture of a book, DVD, CD or videogame cover and get back information such as price, reviews and so on.

Collaborators include Shaolei Feng, C. V. Jawahar, Jiwoon Jeon,
Anand Kumar. Victor Lavrenko and Toni Rath.

Watch the webcast replay >>