ERCIM News No.31 - October 1997

Fourth DELOS Workshop on Image Retrieval

by Carlo Meghini and Pasquale Savino

The fourth workshop of the DELOS working group was on the topic of 'Image Retrieval' and was hosted by CNR in San Miniato, a small medieval town not far from Pisa, from 28-30 August, 1997. DELOS is a Working Group funded by the IT Long Term Research programme of the European Commission to study and investigate existing and emerging technologies and issues relevant to digital libraries.

The DELOS Working Group is just one of a series of ERCIM-sponsored initiatives aimed at promoting research and operational activities in the Digital Library field.

Images are becoming an important medium for information collection and exchange. The amount of images stored and accessed in electronic form is already very large, and is rapidly increasing. The ability to select from a large collection those images that are relevant to a user request, specified either in visual or in textual terms, is at the heart of the image retrieval problem, and, at the same time, a fundamental functionality of a modern information retrieval system. Queries expressed in visual terms (eg, images themselves or elements thereof) are handled on the basis of similarity criteria, and thus aim at providing the user with images that 'look like' the query; this is content-based image retrieval, a form of retrieval currently being studied by the research community. On the other hand, queries expressed in textual terms, such as natural language sentences or formal expressions using a controlled vocabulary with a possibly complex syntax, address the content of the images; this is known as semantic content-based image retrieval. Both these forms of image retrieval were discussed at the Workshop.

The Workshop brought together 20 participants from eight different European countries, as well as 5 invited speakers from the United States, who gave a significantly broad picture of the research on image retrieval under way in the US. The background of the participants ranged from image analysis and processing, to information retrieval and conceptual modelling in the information system area.

In order to stimulate discussion and to support comparative analysis, the Workshop was organized in small sessions, each offering an invited presentation by a well-known expert in the field followed by a presentation from a member of the ERCIM community. An hour was given to the invited speakers, while the ERCIM speakers had about 45 minutes. The discussions following presentations were allowed great temporal liberality to encourage a maximum of spontaneous interaction.

The Workshop began with a comprehensive overview of content-based image retrieval by the first invited speaker, Alberto Del Bimbo from the University of Florence. He presented his work on the employment of image analysis and processing techniques to retrieve images on the basis of colour, shape and textual similarity. Fang Liu from the MIT Media Lab illustrated a number of digital library projects completed or in progress in the Vision and Modeling Group at the MIT Media Laboratory. In particular, she presented a model that can be applied to both spatial and temporal texture features for image retrieval and video analysis. She also showed the results of a scene classification system, which uses both color and texture information to classify indoor and outdoor photographs. David Forsyth from the University of California at Berkeley presented the results of a research project aiming at the retrieval of pictures on the basis of the objects depicted. He first described an approach using segmentation to represent an image in terms of a small number of coherent regions of color and texture. This representation allows the user to view the internal representation of the image and the query results. He then showed how coherent regions can be used to recover people and animals, employing a representation called a body plan. This representation is adapted to segmentation and to recognition in complex environments.

Simone Santini from the University of California at San Diego investigated the problem of meaning in image databases. He argued that the meaning of an image should be considered as a result of the interaction between the user and the database, rather than as an intrinsic property of the images. This leads naturally to replacing the retrieval paradigm with the exploration paradigm. He also presented new concepts and tools (many of them are interface tools) to work in the exploration paradigm. John Smith from the Image and Advanced Television Laboratory of Columbia University presented a new approach for automatically classifying images using image features and related text. The synergy of textual and visual information in Web documents provides a great opportunity for improving the image indexing and searching capabilities of Web image search engines. He presented initial results of classifying photographs and graphics from the Web.

Olivia Frost from the University of Michigan presented a project under way at the University of Michigan to explore the development and evaluation of multimode retrieval schemes which employ both image and text and which are multiple path, iterative and user-directed.

Other presentations from the European researchers ranged from modeling, the study of efficient access methods supporting similarity retrieval, feature extraction methods and applications. In particular, Carlo Meghini (IEI-CNR Pisa) presented an approach for combining syntax and semantics in image retrieval, while Youssef Lahlou (GMD) described an object-oriented model for semantic image indexing. The presentation of Roger Weber (ETH Zurich) and Pavel Zezula (IEI-CNR Pisa) illustrated the problems related to the definition of access structures for similarity retrieval in high dimensional data spaces. Specific solutions were proposed. Roger Mohr (INRIA) presented an approach to image retrieval that uses local characteristics. Ines Oliveira (INESC Lisbon) illustrated a comprehensive set of image processing techniques for video content extraction. Finally, Andre Csillaghy (ETH Zurich) presented an application of image browsing and retrieval applied to astronomical archives.

The next DELOS workshop will address Collaborative Filtering, and will take place in Budapest, Hungary.

Please contact:
Carlo Meghini - IEI-CNR
Tel: +39 50 593405

Pasquale Savino - IEI-CNR
Tel: +39 50 593408

return to the contents page