Association and content-based retrieval

Abstract

In spite of important efforts in content-based indexing and retrieval during these last years, seeking relevant and accurate images remains a very difficult query. In the state-of-the-art approaches, the retrieval task may be efficient for some queries in which the semantic content of the query can be easily translated into visual features. For example, finding images of fires is simple because fires are characterized by specific colors (yellow and red). However, it is not efficient in other application fields in which the semantic content of the query is not easily translated into visual features. For example, finding images of birds during migrations is not easy because the system has to understand the query semantic. In the query, the basic visual features may be useful (a bird is characterized by a texture and a color), but they are not sufficient. What is missing is the generalization capability. Birds during migrations belong to the same repository of birds, so they share common associations among basic features (e.g., textures and colors) that the user cannot specify explicitly. In this paper, we present an approach that discovers hidden associations among features during image indexing. These associations discriminate image repositories. The best associations are selected on the basis of measures of confidence. To reduce the combinatory explosion of associations, because images of the database contain very large numbers of colors and textures, we consider a visual dictionary that group together similar colors and textures. Thus, the visual dictionary summarizes the image features. An algorithm based on a clustering strategy creates the visual dictionary. The associations discovered permit the automatic classification of images during their insertion into image repositories and return accurate and relevant results. More generally, we show that content and knowledge-based indexing and retrieval is more efficient than retrieval approaches based on content exclusively and inaugurate a new generation of approaches in which knowledge contributes to finding images in large image repositories.

Context-sensitive queries for image retrieval in digital libraries

“…Retrieval effectiveness is usually measured in the literature through recall and precision measures (Djeraba 2003). For a given number of retrieved images (the result set rs), the recall R = |rl ∩ rs|/|rl| assesses the ratio between the number of relevant images within rs and the total number of relevant images rl in the collection, while the precision P = |rl ∩ rs|/|rs| provides the ratio between the number of relevant images retrieved and the number of retrieved images.…”

“…"I want to see a portrait," and the portrait example provided to the query engine is chosen to best represent the semantics. The main problem of such approach is that it is not always easy to translate the sematic content of a query in terms of visual features, there is an inherently weak connection between the high-level semantic concepts that humans naturally associate with images and the low-level features that the computer is relying upon (Colombo et al 1999;Djeraba 2003).…”

Integrating Pixel Cluster Indexing, Histogram Intersection And Discrete Wavelet Transform Methods For Color Images Content Based Image Retrieval System

“…color, texture etc) are not powerful enough to specify knowledge queries based on semantics, because the associations between the content and the semantic in the user's mind are weakly declared [17]- [21]. The user has to define the semantic i.e.…”

“…Techniques for retrieval by image semantics are much less developed and also much more difficult problem. Despite some success like automatic scene classification [18], learning semantic association's small images or particular pattern [17] is described.…”

Formalização do domínio imagem para buscas por conteúdo em SGBDs relacionais

“…Em [Brodley et al, 1999] são apresentados testes sobre imagens de CT dc pulmão, com um método que utiliza um conjunto de características que melhor representa imagens já classificadas e outro conjunto de características para classificar subclasses dessas imagens. Já cm [Djeraba, 2003], são utilizados histogramas de cor e descritores de Fourier para textura. Assim, o enfoque é extrair automaticamente a relação entre essas duas características nas diversas imagens.…”