The Meaning of an Image in Content-Based Image Retrieval

One of the major problems in CBIR is the so-called ‘semantic gap’: the difference between low-level features, extracted from images, and the high-level ‘information need’ of the user. The goal of diminishing the semantic gap can be regarded as a quest for similar ‘concepts’ rather than similar features, where a concept is loosely defined as “what words (or images) stand for, signify, or mean” [1]. We first seek to establish a metaphysical basis for CBIR. We look at ontological questions, such as ‘what is similarity?’ and ‘what is an image?’ in the context of CBIR. We will investigate these questions via thought experiments. We will argue that the meaning of an image—the concept it stands for—rests on at least three pillars: what actually can be seen on an image (its ontology), convention and imagination.