Purchase this article with an account.
Krista A. Ehinger, Aude Oliva; Characterizing the shape and texture of natural objects using Active Appearance Models. Journal of Vision 2008;8(6):657. doi: 10.1167/8.6.657.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Previous studies have shown that the way people categorize natural objects is related to the similarity in their visual appearance. Neuroimaging studies, for example, have found that patterns of activation in visual processing regions predict categorization behavior in human observers. In this study, we seek to characterize the space of parameters that observers use to represent the similarity between objects of a single category. We used Active Appearance Models (Cootes, Edwards, & Taylor, 1998) to characterize the visual properties of a category of natural objects (mammals). Active Appearance Models (AAM) describe the exemplars of an image category in terms of shape (defined by a deformable matrix of corresponding image points) and shape-free texture (derived by mapping image texture onto the mean shape). We identified a set of principal components of mammal shapes which are related to human perception of mammals, such as body compactness (which distinguishes tall, long-legged mammals such as deer from small, short-legged animals such as mice), and bulk (which distinguishes heavier mammals such as bears from lighter ones such as dogs). The principal component of shape-free texture in mammals emphasized roughness (smooth skin versus fur) and coloration (darker color on the foreparts than on the hindparts or vice versa). We compare the categorization results from the AAM to categorization patterns found in humans for the same image set and relate the structure of the image space to the structure of human conceptual space.
This PDF is available to Subscribers Only