SPIFF – an AI based service for the image librarian


The image archives of the national library is large and steadily increasing. Currently it is 1.8 million images. A large portion of the images have little or no metada. People maintaining the archives have huge problems in coping with the amounts of information nor do they have tools to do content based search and presentation of images. It is fair to say that we dont know what we have. The AIlab decided to make an open framework for the image archiver’s as a showcase for AI based photo services. This is how the system present itself:

Information about images are to a large extent given on the backside of the image, or in text within the image. In most cases, this is not reflected in the metadata of the image. As a start for a content based image service, we created an AI based OCR service for text extraction from images (front and back). Both handwritten and printed text. This makes the basis for one good strategy for finding images in an archive.

The framework has been built in a modular way to allow for new functionality to be easily attached as it is built. The current version contains functionality for:

For the future we are planning to include functionality for: