返回Google圖書搜尋

Towards Intelligible Query Processing in Relevance Feedback-Based Image Retrieval Systems

出版	INTECH Open Access Publisher, 2008
ISBN	95376190369789537619039
URL	http://books.google.com.hk/books?id=tkmFoAEACAAJ&hl=&source=gbs_api

註釋We have specified within the scope of this paper a framework combining semantics and relational (spatial) characterizations within a coupled architecture in order to address the semantic gap. This framework is instantiated by an operational model based on a sound logic-based formalism, allowing to define a representation for image documents and a matching function to compare index and query structures. We have specified a query framework coupling keyword-based querying with a relevance feedback module managing transparent and penetrable interactions by considering conceptual characterizations of images. The choice of conceptual graphs as an operational model is the most natural in the sense that it holds several advantages in our application context. It indeed allows the symbolic representation of all components of a multimedia indexing and retrieval architecture: queries, index documents and matching function. Moreover its simple representation is particularly well-suited for user interaction in the framework of relevance feedback. To stress the relevance of our approach, the theoretical contributions of this paper in the domain of image indexing and retrieval are summarized below: We have first proposed a neural-network based architecture for the highlighting of image objects, structures abstracting the image visual entites, and the characterization of their associated semantics. In the perspective of unifying the semantic and relational characterizations, we have proposed an integrated model featuring a bi-facetted organization. The visual semantics facet describes the image semantic content and is based on labeling IOs with a semantic concept. The relational facet is itself based on the relational (spatial) characterizations between pairs of image objects obtained after highlighting a correspondence process between extracted low-level information and symbolic relations.