Purchase this article with an account.
Hayaki Banno, Jun Saiki; Superordinate category advantage in scene categorization depends on within- and between-category similarity structure. Journal of Vision 2009;9(8):941. doi: 10.1167/9.8.941.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
What category level is processed fastest when classifying a scene? This is still an open question. Recent studies have shown that Natural vs. Man-made superordinate category classification could be achieved before basic level category. One possibility is that superordinate categories are highly distinctive due to singly sufficient features (Loschky and Larson, 2008; Gosselin and Schyns, 2001). However, we need to take into account the specificity of membership of each category. Distances between basic level categories within a superordinate category are not constant. For instance, City Center scenes seem more similar to Street than Highway, possibly because Highway has strong contrasts to two others in terms of dominant edge orientations (horizontal vs. vertical). If a superordinate category is organized by widely dissimilar categories, the relationship between within-category similarity and between-category dissimilarity would not favor superordinate advantage (Rosch et al., 1976). To test this possibility, we investigated how fast superordinate and basic level categorization were achieved in a go/no-go categorization task. Stimuli were extracted from a database of Oliva and Torralba's (2001) 2 superordinate and 8 basic level categories. In Experiment 1, Natural category was defined as a set of Coast and Mountain, while Man-made as Highway and City Center. In Experiment 2, Highway images were replaced with Street images. Subjects were asked to press a key if briefly presented stimuli (40ms) were targets as quickly and as accurately as possible. In Experiment 1, reaction times of superordinate category classification were not faster than basic level category. Meanwhile, In Experiment 2, speed advantage for superordinate over basic level partially appeared. These results suggest that superordinate categorization is not faster than basic level categorization when membership within a superordinate category are widely distant, and that it is unlikely that scene classification is sequential processing in which either of category levels is processed “first”.
This PDF is available to Subscribers Only