Article published Multimedia
Finding images in large-scale collections
CEA LIST technology was runner-up in the last worldwide assessment of image search engines for the web. Applications range from publishing and audiovisual production to e-commerce and technology watch.
LIST's image search engine method, recognised at the international ImageCLEF assessment, is based on combining linguistic analysis and image processing, simultaneously leveraging all of the information extracted from multimedia documents. "Using semantic resources, we make links between the visual concepts in an image (night/day, inside/outside, urban scene, presence of faces, etc.) and the associated annotations or captions,” explained Olivier Mesnard, leader of the research team. Few laboratories possess this dual expertise, not to mention the ability to combine both areas. The applications are numerous: illustration searches in databases, video searches by content in audiovisual archives (using the automatically transcribed soundtrack), brand or product analyses for automatic marketing watch purposes, and catalogue searches, e.g. for photos of landscapes, monuments or hotels for online travel sales. The team is ready to integrate its technology in any specific application and make improvements, e.g. for real-time analysis.
