Content-based image retrieval

Nowadays the availability of digital cameras allows for the production of thousands and thousands images, either for personal or commercial use. Moreover, national and regional governments are funding projects aimed at preserving the cultural heritage by producing digital versions of documents, pictures, etc. The vast majority of this pictures are stored without any descriptive data associated. However, even if descriptive data si produced with images, still it cannot capture the complex semantic content of images.

To this end, there is a growing need for tools aimed at organising and retrieving images by their semantic content. We address the problem of content-retrieval by a query-by-example approach. In particular, the user is allowed to judge the search results to let the system update the search criteria.

We produced a demo version which is based on some MPEG-7 features and on an original relevance feedback mechanism based on the nearest-neighbor technique. 

A video showing the functionalities of our system can be seen here.

People working on this topic:

• Giorgio Giacinto
• Luca Piras
• Roberto Tronci

Publications on Content-based image retrieval