Semantic modalities in content-based retrieval

This paper analyzes the assumption that the meaning of an image is an inherent property that can be derived from the image data alone, and shows that this assumption is not always adequate. The paper defines three different modalities in which images acquire meaning, and introduces the three query models deriving from them. The first query model is the standard one of databases, and is not considered. The second entails a generalized view of relevance feedback, and the third entails an analysis of the relationship between images and text. These two modalities constitute the focus of the paper