Ideixis --- Image-based Deixis for Recognizing

In this thesis, we describe an approach to recognizing location from camera-equipped mobile devices using image-based web search. This is an image-based deixis capable of pointing at a distant location away from the user's current location. We demonstrate our approach on an application allowing users to browse web pages matching the image of a nearby location. Common image search metrics can match images captured with a camera-equipped mobile device to images found on the World Wide Web. The users can recognize the location if those pages contain information about this location (e.g. name, facts, stories ... etc). Since the amount of information displayable on the device is limited, automatic keyword extraction methods can be applied to help efficiently identify relevant pieces of location information. Searching the entire web can be computationally overwhelming, so we devise a hybrid image-andkeyword searching technique. First, image-search is performed over images and links to their source web pages in a database that indexes only a small fraction of the web. Then, relevant keywords on these web pages are automatically identified and submitted to an existing text-based search engine (e.g. Google) that indexes a much larger portion of the web. Finally, the resulting image set is filtered to retain images close to the original query in terms of visual similarity. It is thus possible to efficiently search hundreds of millions of images that are not only textually related but also visually relevant. Thesis Supervisor: Trevor Darrell Title: Professor of Computer Science