Purchase this article with an account.
Ilan Kadar, Ohad Ben-Shahar; A new perceptual paradigm and psychophysical evidence for hierarchical gist recognition. Journal of Vision 2011;11(11):1113. doi: https://doi.org/10.1167/11.11.1113.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The “gist” of a visual scene is often synonymous with its basic-level category (e.g. coast, street) and the remarkable ability of humans to recognize it both rapidly and accurately is highly useful in everyday life. Following the seminal studies by Rosch et al. (1976) and Tversky & Hemmenway (1983), it has been assumed that basic-level categorization is privileged over the superordinate-level (i.e., indoor vs. outdoor) because it maximizes both within-category similarity and between-category variance. However, recent research has begun to challenge this view (e.g., Fei-Fei et al., 2007; Loschky & Larson, 2008, 2010). Here we make study these directions more fundamentally by investigating the perceptual relations between scene categories to determine whether or not gist recognition is a hierarchical process, and if so, what hierarchical structure does it exhibit in humans.
We introduce a novel psychophysical experimental paradigm - the category discrimination paradigm - where we briefly present two real-world scene stimuli simultaneously for different presentation times and ask subjects to respond whether they belong to the same basic-level category or not. As we show, proper analysis of the obtained data can reveal hierarchical perceptual distance between different scene categories and a corresponding hierarchical structure at the perceptual processing level. In particular, we show that the decision whether the scene is man-made or natural is made first, and only then followed by more complicated decisions (such as whether a man-made scene is indoor or outdoor), when fewer candidate categories are still viable. We argue that this observed hierarchical structure not only improves performance, but is also faster to execute in both biological or artificial visual systems.
This PDF is available to Subscribers Only