What makes a difference: A method to determine whether a change in an image affects the perceived gist

摘要

Although many changes in images and real-world situations can go undetected, researchers have suggested that changes affecting the gist of an image are likely to be noticed. This article proposes a method to determine whether a change in an image affects its gist. The method was designed to determine a systematical difference between sets of descriptions of subjectively interpretable images. First, for each image, a set of descriptions is generated. Then, a rater judgesgood fit orpoor fit for the descriptions for each image, providing the base for categorization. Finally, a Fisher exact probability test is used to determine a difference between sets of descriptions. The method is applied to determine a gist change for 18 test sets. The proposed method makes it possible to compare subjective interpretations of images, while leaving room for differences in interpretation of an image by different people.