Document Actions
Lighting Invariance
Our visual system has the amazing ability to recognize thousands of individual faces, animals, and objects, successfully discounting irrelevant changes and using relevant ones to distinguish among known individual instances. How is this invariant representation for an individual acquired? What neural changes occur in the brain as invariance is learned? Researchers have investigated invariance in a number of domains: viewpoint invariance where recognition occurs despite rotations of the object, position invariance despite translations of the object, size invariance despite scaling of the object, etc. Here, we tackle the question using lighting invariance. Why? It is the one domain of invariance that has a computational framework for generalizing across lighting conditions after a few exemplars.
Perceivers can easily recognize objects in a variety of lighting environments; however, these lighting variations cause complex alterations of the perceived low-level features of the object. Tarr and colleagues (1998) argue that lighting can alter at least three attributes: the overall magnitude of illumination throughout the image; the shading information that depends on the interaction of the shape of the object and the light source; and cast shadows reflected by the object on itself that are based on surrounding surfaces. To successfully recognize objects, the visual system must have a mechanism to separate variations due to the lighting source from those due to different objects. Models of object recognition typically propose one of two solutions: a memory storage that is independent of the shading and shadow information or a storage that encodes the effects of the illumination on the object. Consequently, a prevalent experimental manipulation is to train participants to recognize novel objects that are illuminated from a particular lighting direction. Following successful identification of the learned stimuli, participants are then tested with the same objects under different lighting conditions. If only the underlying structure of the object is stored, the change in illumination should have no effect on the recognition response time or accuracy of the same object in a new lighting condition. Alternatively, if the illumination effects on the object are stored, the change in illumination should make it more difficult to recognize the object.
Methods.
The first challenge is to identify a behavioral paradigm that captures the acquisition of a representation of face identity that is invariant to changes in lighting. Once a behavioral correlate is found, an imaging experiment will examine the neural substrate of the invariance.
Results and Discussion.
The first challenge has turned out to be tricker than first presumed. Several behavioral studies have been run, but the paradigm is not refined enough yet. Stay tuned...
Current and future work.
See above! :)
People. Jean
Congrats to Jessie and Nigel!