SEMCOG: a hybrid object-based image database system and its modeling, language, and query processing

Abstract

Image data is structurally more complex than traditional types of data. An image can be viewed as a compound object containing many sub-objects. Each sub-object corresponds to image regions that are visually and semantically meaningful (e.g. car, man, etc.). We introduce a hierarchical structure for image modeling that supports image retrieval, at both the whole-image and object levels, using combinations of semantic expressions and visual examples. We introduce an image database system called SEMCOG (SEMantics and COGnition-based image retrieval). SEMCOG aims at integrating semantics- and cognition-based approaches and allows queries based on object-level information. We present a formal definition of a multimedia query language, we give details of the database's implementation and query processing, and we discuss our methods for merging similarities from different types of query criteria.