Semantic Intensity: a measure of visual relevance

This work deals with the quantification of the effects of discarding large amounts of irrelevant information in sets of images. In particular, the trade-off between the cost of losing information and the benefit of increasing perceptual speed is examined. We look for a measure of such trade-off by a new measure, 'semantic intensity', broadly defined as the ratio of amount of 'meaning' conveyed to the amount of 'effort' required to perceive it. The establishment of more precise definitions of these two quantities ('meaning' and 'effort of perception of meaning'') is the pervasive theme of the paper. A preliminary operational definition of semantic intensity is introduced and applied to a comparison between avatar and semantroid showing the advantage of the latter. A more precise definition of semantic intensity is then given by relating 'meaning' to information and 'effort' to bit geometric density, compactness, contrast and homogeneity. The relation of 'meaning' to information is mediated by an extension of Relevance Theory from the verbal to the visual domain, resulting in a new constraint which qualifies information as 'meaning'. The constraint is the requirement of direct representation, or pointing to a representation, without intermediate mental operations. Application of this new definition of semantic intensity also shows the advantage of semantroids over avatars, in many practical cases. This work may be regarded as a first step toward a quantitative extension of Relevance Theory to the visual domain.