Abstract:
Whether we try to find a hidden cat, or detect people crossing the street at night - poor contrast or occluding items often render objects partly invisible. We thus have to interpolate some structures while ignore others to perceive objects. How does our visual system use stimuli like real contours (RCs) to construct object outlines from illusory contours (ICs)? To study this, I psychophysically measured the perceptual strength of ICs with different contextual RCs in human subjects.
Can RCs interact with ICs, or do they interact with inducing processes? In an orientation discrimination task, an abutting line IC was presented for different durations, immediately followed by a backward mask consisting of RCs. These were either parallel to IC or inducers, or orthogonal to either of them. RCs are masked by parallel lines (Li 2000). RCs parallel to the inducers therefore mask the inducers. RCs parallel to the IC, however, might mask the IC. As IC processing takes longer than RC processing (Lee 2001), inducer masking should occur at shorter presentation times than IC masking. Experiments supported this: masks parallel to the inducers interfered early (<100 msec), while masks parallel to the IC interfered later (>125 msec). Thus, RCs can interact directly with either IC-induction or ICs.
Subthreshold contrasts are not perceived, yet can be processed by V1. If ICs are processed in V1, subthreshold RCs might interact with them. Curvature discrimination thresholds of Kanizsa-figures were measured with superimposed parallel or orthogonal RCs of different contrasts. Parallel RCs were found to interfere at suprathreshold, orthogonal lines interfered at subthreshold. Thus, RC-IC interaction occurs even at subthreshold, while its orientation-dependency changes with contrast. This suggests either receptive field changes at subthreshold (Polat 1998, Mareschal 2002), or an underlying orientation-reversed feedback mechanism (Ramsden 2001).
RC-IC interactions in the proposed mechanism would change orientation dependency over time. This was tested with curvature discrimination of Kanizsa-figure ICs, as described above. Superimposed RCs of different contrasts were presented at varying stimulus onset latencies. RCs supporting the IC percept early on (<100 msec) interfered later (>100 msec) and vice versa. Data thus showed an orientation-reversal over time, which was most prominent at subthreshold. These results are consistent with a V1-V2 feedback system in IC processing.
Are contextual effects on IC processing evident in V1? This was tested using single-cell recording in the macaque. Moving Kanizsa-figures were displayed with and without superimposed parallel or orthogonal RCs. V1 neurons responded stronger to nonoptimally oriented RCs when presented abutting to an IC than when presented with a control (rotated inducers). Responses to optimally oriented RCs were not different in IC and control stimulus. These preliminary results indicate that V1 neural reponses to RCs can be changed by IC processing.
The perception of context effects on ICs is thus consistent with an orientation-reversed feedforward-feedback mechanism in areas V1 and V2. To reconcile the psychophysical results with the preliminary physiological findings, however, further physiological studies are necessary.