Object boundaries in the natural environment are often defined by
changes in luminance; in other cases, however, there may be no difference
in average luminance across the boundary, which is instead defined by more
subtle 'second-order' cues, such as changes in the contrast of a fine-grained
texture. The detection of luminance boundaries may be readily explained in
terms of visual cortical neurons, which compute the linear sum of the excitatory
and inhibitory inputs to different parts of their receptive field. The detection
of second-order stimuli is less well understood, but is thought to involve
a separate nonlinear processing stream, in which boundary detectors would
receive inputs from many smaller subunits. To address this, we have examined
the properties of cortical neurons which respond to both first- and second-order
stimuli. We show that the inputs to these neurons are also oriented, but with
no fixed orientational relationship to the neurons they subserve. Our results
suggest a flexible mechanism by which the visual cortex can detect object
boundaries regardless of whether they are defined by luminance or texture.
