The evolution of visual information retrieval

Abstract

This paper seeks to provide a brief overview of those developments which have taken the theory and practice of image and video retrieval into the digital age. Drawing on a voluminous literature, the context in which visual information retrieval takes place is followed by a consideration of the conceptual and practical challenges posed by the representation and recovery of visual material on the basis of its semantic content. An historical account of research endeavours in content-based retrieval, directed towards the automation of these operations in digital image scenarios, provides the main thrust of the paper. Finally, a look forwards locates visual information retrieval research within the wider context of content-based multimedia retrieval. Keywords: visual information retrieval; image retrieval; video retrieval; semantic image retrieval; content-based retrieval The context of visual information retrievalThe retrieval of images or image sequences that are relevant to a query is a long-established activity which has evolved quite remarkably during the last 50 years, from the special preserve of a relatively few professional practitioners to the forefront of research in computer vision and a leading edge domestic application of information technology. This extension of traditional information retrieval activity includes both still and moving images, the former usually characterized in the literature as 'image retrieval', the latter as 'video retrieval', and the two in combination, sometimes, as 'visual information retrieval' [1,2].The literature of visual information retrieval has grown at a stupendous rate. To quote Jörgensen, in her landmark text within the field: Adjectives such as 'vast' are often applied to the various literatures … related to image processing, but even this designation is an understatement [3: p. 199].More remarkable still is the fact that almost all of that growth has taken place since the early 1990s, and reflects those technological advances which brought the digital image to the attention of the computer scientist. Greatly increased availability of images via the Internet, then via mobile platforms, and most recently as an aspect of the social networking phenomenon, has been said to place us

Knowledge Organisation and its Role in Multimedia Information Retrieval

“…techniques specifically for image retrieval as a technology which can be applied to multimedia IR. Enser (2008a) relates the failure of CBIR systems based on A.I. techniques to fulfil their promise, with critics in the information science community demonstrating through experimentation that users do not find low level features useful for search.…”

“…Commercial systems that proposed using such an approach have failed to make any headway as a result. A focus on the more semantic aspects of the content (Enser, 2008a) has proved to be problematic also as some concepts in an image are intrinsic to it and are not physically present (e.g. a picture of a politician involved in an election -the 'politician' is identifiable, but the concept of an 'election' is more difficult to detect).…”

Les éléments formels de la fiche descriptive des banques d’images et leur nécessaire normalisation

“…La recherche par le contenu est un champ de recherche scientifique très productif qui crée des méca-nismes pour récupérer de grandes collections d'images (Tsai, McGarry et Tait, 2006). Ce champ de recherche commence à se développer à la fin des années 1970, sous la forme de bases de données construites spécialement pour le stockage et le repérage d'images (Enser, 2008). Le paradigme de recherche et de repérage d'images par contenu fonctionne à partir du contenu explicite de l'image numérisée, c'est-à-dire à partir des pixels.…”

“…Le paradigme de recherche et de repérage d'images par contenu fonctionne à partir du contenu explicite de l'image numérisée, c'est-à-dire à partir des pixels. Certains auteurs se montrent sceptiques quant au choix de ce terme, puisqu'il peut prêter à confusion avec les aspects sémantiques du processus (Enser, 2008). Ce type de repérage utilise les métadonnées caractérisant les formes, les couleurs et les textures de l'image.…”

“…Ce type de recherche et de repérage d'images est discuté dans la littérature scientifique à partir du milieu des années 1990 (Enser, 2008). Une partie de la recherche doit se réaliser verbalement (Enser, 2000) ; on utilise les mots-clés attribués aux images ou bien on recherche en mode plein texte.…”