Broadening the horizons of scene gist recognition: aerial and ground-based views

Loschky, Lester and Ellis, Katrina and Sears, Tannis and Ringer, Ryan and Davis, Joshua (2010) Broadening the horizons of scene gist recognition: aerial and ground-based views. Journal of Vision, 10 (7). p. 1238. ISSN 1534-7362

Full content URL: http://dx.doi.org/10.1167/10.7.1238

Full text not available from this repository.

Item Type:Article
Item Status:Live Archive

Abstract

Numerous studies in the last decade have used ground-based views of scenes to investigate the process of scene gist recognition. Conversely, few if any studies have investigated scene gist recognition of aerial (i.e., satellite) views. This study asks the question, how much of what we know about scene gist recognition from ground-based views directly translates to aerial views?

Fifty-two participants were randomly assigned to Aerial and Ground-based conditions, with processing times (SOA) and scene categories varied within-subjects. Stimuli were monochrome photographs from 10 categories: 5 Natural: coast, desert, forest, mountain, river; 5 Man-made: airport, city, golf-course, residential, stadium. Aerial images were from Google Earth©. Both target and mask images were presented for 24 ms, with SOAs of 24-94 ms plus a no-mask condition. Participants then chose between all 10 categories.

As predicted, ground-based views were recognized more accurately than aerial views. However, contrary to predictions, aerial view recognition did not benefit more from additional processing time than ground-based view recognition. Aerial view performance with no mask was worse than ground-based view performance at 24 ms SOA. Thus, gist perception of aerial views is more data (information) limited than resource (time) limited, perhaps because they are “accidental views” (Biederman, 1987). An additional analysis collapsed all 10 basic level categories into 2 superordinate level “Natural” and “Man-made” categories. For ground-based views, Natural categories were consistently high, whereas Man-made categories benefited from additional processing time. However, for aerial views, both Natural and Man-made categories benefited equally from additional processing time. Nevertheless, confusion matrices for the 10 basic level categories and responses showed a correlation of .80 across the Aerial and Ground-based views, suggesting that discriminability between categories is similar across aerial and ground-based views. Further research will investigate what information both aerial and ground-based views contain, and what information aerial views lack.

Additional Information:Numerous studies in the last decade have used ground-based views of scenes to investigate the process of scene gist recognition. Conversely, few if any studies have investigated scene gist recognition of aerial (i.e., satellite) views. This study asks the question, how much of what we know about scene gist recognition from ground-based views directly translates to aerial views? Fifty-two participants were randomly assigned to Aerial and Ground-based conditions, with processing times (SOA) and scene categories varied within-subjects. Stimuli were monochrome photographs from 10 categories: 5 Natural: coast, desert, forest, mountain, river; 5 Man-made: airport, city, golf-course, residential, stadium. Aerial images were from Google Earth©. Both target and mask images were presented for 24 ms, with SOAs of 24-94 ms plus a no-mask condition. Participants then chose between all 10 categories. As predicted, ground-based views were recognized more accurately than aerial views. However, contrary to predictions, aerial view recognition did not benefit more from additional processing time than ground-based view recognition. Aerial view performance with no mask was worse than ground-based view performance at 24 ms SOA. Thus, gist perception of aerial views is more data (information) limited than resource (time) limited, perhaps because they are “accidental views” (Biederman, 1987). An additional analysis collapsed all 10 basic level categories into 2 superordinate level “Natural” and “Man-made” categories. For ground-based views, Natural categories were consistently high, whereas Man-made categories benefited from additional processing time. However, for aerial views, both Natural and Man-made categories benefited equally from additional processing time. Nevertheless, confusion matrices for the 10 basic level categories and responses showed a correlation of .80 across the Aerial and Ground-based views, suggesting that discriminability between categories is similar across aerial and ground-based views. Further research will investigate what information both aerial and ground-based views contain, and what information aerial views lack.
Keywords:Scene Gist
Subjects:C Biological Sciences > C890 Psychology not elsewhere classified
Divisions:College of Social Science > School of Psychology
ID Code:7550
Deposited On:17 Feb 2013 17:31

Repository Staff Only: item control page