"Picture This--Automatic Image Categorization"

"Picture This--Automatic Image Categorization"
IST Results (05/02/05)

The IST-funded LAVA project has found a way to automatically categorize or classify visual images without the use of additional metadata by bringing together researchers in the fields of machine-learning and computer vision and cognitive science, according to LAVA participant Gabriela Csurka with Xerox Research Center Europe. "We began our approach by grouping together similar types of objects...and trying to find a way of categorizing those that were common to a group," she explains. "We applied machine-learning techniques to find the distinctions between images by focusing on sections of images that were similar--sections that were common to other images with the same content." Another challenge for the LAVA team was to find a way to accurately categorize image content in spite of views of objects taken from different distances or perspectives. The researchers have successfully devised a technique for capturing visual images and automatically identifying the proper category those images belong to, and they think this advance has dramatically augmented the ability to create reliable vision-based detectors for common events and objects. The LAVA technologies' potential applications include image browsing within documents, consumer photo archival and image management, Web-based image searches, human-computer interaction, video surveillance, robotics, and medical imaging. The LAVA team won 14 out of 18 detection, localization, and classification contests in the PASCAL network's Visual Object Classes Challenge to amass a standardized group of object recognition databases and supply a common set of tools for accessing and managing annotations in those databases.

http://istresults.cordis.lu/index.cfm/section/news/
tpl/article/BrowsingType/Features/ID/75489