Pictures may 'tell a thousand words,' but in today's megapixelated world of digital images, sorting and identifying them will be made much simpler through the work of EECS assistant professor Antonio Torralba and colleagues in the Computer Science and Artificial Intelligence Lab, CSAIL.
Torralba and his team have been developing very short codes or numerical representations that can be derived from individual images to enable automated cataloging of the billions of images on the Internet. Current to future applications of this work range from automatic indexing of digital images through downloadable software to making true machine vision possible in the future enabling robots to make sense of visual (numeric) data from their cameras and use this to locate themselves.
Specifically, Torralba's image data system will provide representation of a set of 12.9 million images from the Internet with just 600 megabytes of data--now available online and small enough to fit in the RAM memory of most current PCs and/or stored on a memory stick. In addition the software to enable searches of the database are being made universally available online. And the searches can be done in less than a second!
The key to Torralba's system is keeping the amount of data per image extremely small--a feature that does not prevent recognition and sorting--yet speeds the process. In addition, the system does not require break down of images into discrete sections as do many other current methods.
Torralba will be presenting his latest findings next month at the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, CVPR 2008, being held in Alaska. He has collaborated in this work with Rob Fergus at the Courant Institute in New York University and Yair Weiss from Hebrew University in Jerusalem. Torralba's work has been supported in part by a grant from the National Science Foundation.
Read more about this work in the MIT News Office article of May 21, 2008.