Researchers at Xerox Research Centre Europe (XRCE) have demonstrated a software technology that can link text and general images together - a breakthrough in how online and paper-based information is categorized.
Current tools classify or "tag" either text or images so they can be processed; but until now no one has combined the two effectively, according to computer scientist Marco Bressan who led the research team. By linking image and text-based content, this new software technology significantly improves fundamental document management tasks like retrieving information from a database or automatically routing documents. The result? More complete searches and streamlined business processes.
For example, if a brochure from an isolated hotel in the French Alps describes the hotel's features and includes maps and pictures of mountainous surroundings, the categorizer will automatically discover the content and link the text and the images together. Then someone searching for an isolated mountain lodge within a certain price range would retrieve the brochure even if "isolated lodge in the mountains" were never mentioned in the actual text.
The research aligns with Xerox's goal of developing smarter documents to make information-based work easier, more efficient and more effective. Bressan believes there are many uses for the new categorization software.
"Suppose a traveler wants to combine vacation photos with a journal to produce an annotated photo album or photoblog recapping vacation highlights," said Bressan. "Because the Xerox categorizer handles both text and visuals, it can identify the photos, automatically match them to the written text and then enrich the visuals with additional information via hyperlinks to a knowledge base such as Wikipedia."
A second application, according to Bressan, could be at Xerox's imaging centers, where the company scans and digitizes documents to create secure, accessible and searchable online information archives for its customers. Currently the process of scanning, labeling and indexing documents is partially supervised by operators. Hybrid categorization can streamline document management in this application, improving accuracy and eliminating manual operations.
Enabling Xerox's hybrid categorizer are recent advances in machine learning and pattern recognition, advances in computer vision and the large body of hybrid content now available. XRCE has extensive experience with text categorization and, in 2005, demonstrated the industry's first generic image categorizer. The new categorizer combines earlier text and image categorizers to handle hybrid content, with powerful results.
"Xerox's hybrid categorizer creates a shared knowledge space between text and images," said Bressan. "The textual information enriches the visual, and the visual information enriches the textual. The whole is ultimately greater than the sum of the parts."