Modeling and Information Processing

INDEX - Image indexing

Objective

• To design semi-automatic methods for indexing large image type databases by keywords.

Onera's contribution

In the current state of the art, automatic indexing of images or videos by keywords is impossible and will probably always remain so given the strongly polysemantic character of multimedia contents. Our approach, based on automatic learning techniques and the loopback of pertinence consists of helping the human operator responsible for the indexing operation by proposing several choices of sets of keywords. The useable keywords belong to an ontology related to the area of application.

In its incremental learning, the indexing support system uses the image's characteristics, which may be its radiometry, shapes or composition, to propose indexing choices to the operator. This approach gets around the problems of polysemantics and extracting characteristics (weakness of the automatic segmentation techniques).

Illustrations

• C4ISR: indexing aerial/satellite images, multimedia indexing for intelligence
• Semantic web: search engine for images and videos by contents
• Infomagic: participation in the IMVN competitiveness center [center dedicated to the development of new contents, intended for six main sectors of activity: cinema and audiovisual, computer games, digital heritage, knowledge engineering, education, communication].

Last Update: 11 October 2006 - © ONERA 2009 - Terms of use