Overall Objectives
Scientific Foundations
Application Domains
New Results
Contracts and Grants with Industry
Other Grants and Activities

Section: Application Domains

Keywords : Image Databases, Photo Agencies, Digital Pictures, Medical Imagery, Text-Image Indexing.

Still Image Database Management

We are particularly interested in large image bases, like those managed by photo agencies. These agencies have between five hundred thousands and twelve millions of images. The Andia Press agency has a million of images, Sigma twelve millions, the Corbis agency which gathers the whole Bill Gate's acquisitions has thirty six millions of images. These agencies work according to two modes. In the first one, they respond to a customer's query by sending him/her a set of images. The customer pays for the images that he/she publishes. In the second mode, the customers are subscribers ant the agencies send them their new photographs, the mode of payment being identical. This working method is that of the AFP or Reuters.

One of the concerns of the agencies is of course the digital rights management, and the fact that they are not unduly used by people or institutions while not having discharged the rights. Watermarking and indexing are two techniques planned to control image diffusion, either by seeking a watermark of property in the images, or by checking by indexing that the image is not a fragment of an image of the agency base.

Another important field where the management of the images acquires an increasing importance is that of the medical images. The access to the medically interesting contents of the image is a true difficulty, so is the level of quality imposed by this field to the recognition system. The applications of content-based methods are thus still to come in this field.

Traditional image indexing consists in automatically extracting from an image numerical descriptors representing the color, texture, interest points or other similar information. However, such descriptors are not relevant to tackle the problem of a ``semantic'' querying of an image database: how can a customer find the pictures of a sunset or the pictures of his/her daughter learning to swim? How can an archivist in a news agency find a relevant picture to illustrate an article dealing with poverty in India? One way to address this problem is to make the most of existing documents in which images and texts appear together, and then use relevant parts of the texts to index the images; nonetheless, the definition of such text-image indexing schemes are up to now under study.


Logo Inria