Researchers Patent Image Recognition Method
Posted: April 13, 2007 at 1:00 am, Last Updated: November 30, -0001 at 12:00 am
By Dave Andrews
Two professors and a former graduate student in the Volgenau School of Information Technology and Engineering have patented a method for “fingerprinting” and recognizing images.
The patented invention is the work of Sushil Jajodia, director of the Center for Secure Information Systems; Zoran Duric, associate professor of computer science; and Neil Johnson, a former graduate student.
The Fingerprinting and Recognition of Data invention addresses many problems in the field of image recognition, especially when searching for an image via the Internet where many images are illegal copies and distortions. The invention systematically tracks these images and is capable of filtering through large repositories.
“This patent isn’t just important to us as researchers. It is useful to anyone who has a need to provide copyright protection to multimedia objects such as images and videos,” Jajodia says.
The current process of identifying illegal copies of copyrighted information relies upon digital watermarking techniques as well as a database. The invention offers a novel solution to the many other classification methods that are time consuming and error prone.
“An advantage of the invention is that the fingerprint recognition process is fast and reliable,” Jajodia explains. “It also generates fingerprints that are capable of surviving image distortions, such as cropping, affine transformations, color shifts, blurring, resealing and resampling. Additionally, it may survive manipulation by tools designed to disable watermarking techniques.”
Unlike digital watermarking, in which the user can search for a specific embedded image code to yield that specific image, this invention uses fingerprints that can match similar images. The fingerprints contain information about the visual features of the image. These features may be efficiently sized so that they may be easily searched using standard database technologies.
Images may then be indexed or categorized using the visual features in the fingerprint as well as text annotation, assigned subjects or image types.
The invention also addresses the problems of current histograms used to compare images. The inventors assert that color histograms lack spatial information, making images with very different appearances seem similar. They also say that various image databases assume that all images are scaled to the same amount of pixels. The invention is capable of recognizing such distortions and distinguishing between the images.
Jajodia and Duric say the device has yet to be developed into a product, but they are interested in licensing the technology for development.