Free sorting of images: Attributes used for categorization

Abstract

Categorization is crucial to human perception and cognition. Without the abilities to categorize, people would not be able to deal with the multitude of things they perceive daily. Past research has shown that categorization also facilitates effective and efficient browsing of both text and image documents especially when users have no specific document in mind but only some idea of what they are looking for. However, the question of what types of attributes should be used to index categorieslgroupings of images is not fully answered. Knowledge about how people categorize images and assign names to categories may provide some clues to help answer this question. Results of this free sorting study, as well as other research on the nature of categorization and search requests by image users, point to the fact that interpretive attributes are better candidates than perceptual attributes for indexing categorieslgroupings of images.

Image semantics in the description and categorization of journalistic photographs

“…This is consistent with the results from several earlier studies (Fidel, 1997;Greisdorf & O'Connor, 2002;Jörgensen, 1998;Jörgensen, 1999;Mojsilovic & Rogowitz, 2001a;Rorissa & Hastings, 2004). High-level and affective concepts were also used in image categorization while some but relatively few categories were based on similar visual elements.…”

“…The affective responses to images and the atmosphere of the image itself have been deemed important in the selection process of journalistic photographs (Kobré, 2000;Markkula & Sormunen, 1998) and based on these categorization results they also seem to be a natural way for people to organize photographs. The influence of the photographs' emotional tone on categorization has not been discussed much in previous studies, although the image categories reported by Rorissa and Hastings (2004) included several affective themes. Other studies on free categorization (Vailaya et al, 1998;Teeselink et al, 2000) seem to concentrate purely on the generic semantic level of image content (justified because of their interest in, for example, automatic image classification), whereas several of the journalistic photographs used in this study contained affective subject matter.…”

“…The second most important feature was whether the photographs depicted a landscape or objects/buildings. Based on their recent study on free categorization of images, Rorissa and Hastings (2004) concluded that interpretive attributes are better candidates than perceptual attributes for indexing categories of images. In their experiment the main categories formed by the subjects were: exercising, single men/women, working/busy, couples, poses, entertainment/fun, costume and facial expression.…”

“…The percent overlap gives a measure of similarity, which was then converted to a measure of dissimilarity: δ ij = 1 -S ij . Rorissa and Hastings (2004) also used this procedure in their study of free sorting of images. The hierarchical cluster analysis was done using the complete-linkage (farthest-neighbor) method, applied in similar studies also by Lohse et al (1994), Vailaya et al (1998) and Teeselink et al (2000).…”

“…For this reason the hierarchical cluster analysis was carried out separately for the two groups of subjects and Jaccard's coefficient was used to measure if the clustering results differed between the groups. It was also used by Lohse et al (1994) and Rorissa and Hastings (2004) to test the consistency of subjects in sorting tasks. The calculated Jaccard's coefficient value for the two subject groups of this study was 0.788.…”