Originating from a Dagstuhl seminar in 1999, this is a collection of fifteen articles which cover various aspects of content-oriented image and video retrieval using automatically derived features such as the images’ color, texture, and shape information. Five papers deal with image retrieval – from the point of view of perceptually relevant features, the derivation of local image features from affinely invariant regions, a very efficient method of invariant feature extraction using a Monte Carlo estimation method, 2D shape-based object retrieval using the best possible correspondence between visual parts, as well as an overview of features used in 44 contemporary image retrieval systems. Methodological problems are focused on in papers dealing with a probabilistic approach to object recognition and localization in images, distribution-based measures for image similarity and their application for image retrieval, non-parametric density estimation of image feature clustering, and the representation of semantic information in a probabilistic object-oriented logic within a prototype multimedia retrieval system.
Problem of video retrieval are discussed in a dynamic content description approach to video sequences, including a hierarchical motion classification for video, the derivation of facial and motion features from optic flow, the hierarchical clustering of video key frames and its use for video summarization based on an asymmetric measure for image similarity, and the extraction of semantic information from lower-level features and signs based on semiotic considerations. Finally, two papers deal with similarity search in large image databases that is adaptable to application-specific requirements and user preferences, and parallel nearest neighbor searching in large multimedia repositories on a network of workstations.