A method of measuring the semantic gap in image retrieval: Using the information theory

Abstract

The semantic gap exists in content-based image retrieval. Many researchers have proposed a variety of methods to bridge or narrow this gap. The methods include into two types: bottom-up and top-down approaches. These approaches have made great progress, but few studies have been done in how to measure it. In this paper, we redefine the semantic gap in a user-centered way and present a method for measuring the semantic gap, using the information theory.