Still, J. D., Dark, V. J. & Parkhurst, D. J. (2008). Dense Regions of View-Invariant Features Promote Object Recognition [abstract]. Poster presented at the 16th annual meeting of the Object Perception, Attention & Memory Conference, Chicago, IL.

Using SIFT (Scale Invariant Feature Transform), a computer vision algorithm, Still, Dark, and Parkhurst (2007) showed that eye fixations are linked to regions containing high proportions of SIFT view-invariant features. We examined whether regions of invariance affect object naming accuracy. Participants named fragmented images displaying 20%, 40%, 60%, or 80% of the pixels of photographs of complex objects. Displayed pixels were chosen based either on random or view-invariant features. Object naming accuracy was better when regions of invariance were shown, especially with 40% or fewer pixels. Regions of invariance appear to play a role in human object recognition.

<-- back