Walker, L.L., Perttula, D.J. & Malik, J. (2000). Top-down Influences on Grouping in Naturalistic Stimuli. Invest Opthalmol Vis Sci, 41(4), 2316.
PURPOSE: We have developed an apparent motion paradigm that can be used to determine perceptual groups in natural images. Peterson (1994) has shown that object recognition influences figure-ground organization using ambiguous figures. The goal of this work was to study the analogous question for grouping in natural images. The Gestaltists and researchers such as Kanizsa argue that grouping is bottom-up, while examples like the Dalmatian dog images have been used in favor of a significant top-down role. Preliminary results with our apparent motion paradigm demonstrate that object recognition is not required for low-level grouping. METHODS: Subjects are presented with gray-scale photographs for 200ms, followed by the occluding contour of a group in the photograph for an apparent motion effect. This effect does not occur for non-groups in the scene. The 2AFC task requires the subject to respond whether the contour shifted to the right or left. The images subtended a visual angle of approximately 6°x9.5° and were drawn from the Corel image database. Typical examples include photographs of animals in their natural habitats. To test whether object recognition was needed for grouping and resulting apparent motion, subjects were presented with variants of the images to impede recognition. Variants included inversed images, flipped images, degraded images, and flipped degraded images. If object recognition is required for grouping perception, performance should decrease for the images variants, resulting in longer response times and possibly a decrease in percent correct. RESULTS: Normal image presentations and image variants produced equivalent performance across subjects. On average, subjects responded with the motion direction correctly around the 90% level. Response times only varied by 1.5% between all image conditions. CONCLUSIONS: Object recognition has no significant effect on grouping perception in this apparent motion paradigm, using naturalistic stimuli. If our conclusions are correct, this paradigm supports the classic argument that bottom-up inputs are sufficient for low-level grouping perception.