
Lights, camera, action! An interaction between illumination and viewpoint change in object recognition
Author(s) -
Wendy D. Zosh,
Quoc C. Vuong,
Michael J. Tarr
Publication year - 2010
Publication title -
journal of vision
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.126
H-Index - 113
ISSN - 1534-7362
DOI - 10.1167/2.7.40
Subject(s) - observer (physics) , computer vision , artificial intelligence , computer science , object (grammar) , visual field , orientation (vector space) , mathematics , physics , optics , geometry , quantum mechanics
Research in spatial cognition and object recognition has indicated that an “active” observer (i.e. moving) shows an advantage in their ability to recognize an object from a different viewpoint relative to a “passive” observer (i.e. stationary) who is presented with the same image geometry. Some researchers have attributed this advantage to the contributions made by the body senses (e.g. vestibular and proprioceptive) to an observer's ability to spatially update their location in the environment. However, a potential source of information that may be exploited by the visual system is the differential effect an interaction between illumination and viewpoint has for observer and object movement. Phenomenologically, the retinal projections will differ for the two types of motion due to the interaction with illumination sources. If an object rotates relative to a fixed light source, approximately the same area of the visual field will be illuminated, whereas when an observer moves about an object (relative to a fixed light source), the area illuminated in the visual field will change with orientation. The overall pattern of shading and shadows will differ for the two conditions despite equivalent physical geometries across rotations. To address this, we investigated whether the interaction between illumination and viewpoint change provides sufficient visual information to confer the same advantage seen for an active observer to a stationary observer. Preliminary findings suggest that the local feature information contained in images is sufficient to show an active observer advantage in object recognition