A DEEP LEARNING APPROACH TO ENRICH PHOTOGRAPHIC ARCHIVES

Photographic archives are the storehouses of humankind's visual memory, and the documents stored in them form a cultural testimony asset of incalculable value. Their main mission is to preserve their cultural assets and to disseminate them to the rest of the society. From all the documents stored and managed in historical archives an average of 90% correspond to photographs that have an enormous cultural and historical value. They are snapshots of our society's past, and they enable us to see the day to day life of our ancestors.

Deeparchive is a project aiming at direct technology transfer in two key areas of photographic archives' activity: cataloguing, and access provision. While the archivist community has been very active in the early implantation of new information technologies, photographic archives still rely on manual annotation of their assets and traditional keyword-based search engines with several known limitations. With this project we propose to leverage the recent advances in Machine Learning and Computer Vision research to: (1) create new tools to help archivists in the tasks of classification, annotation, and indexation of new data; and (2) propose a transition from the keyword-based to the semantic-based search paradigm in photo archives' access.