Abstract
Recent studies have shown that visual perception is serially dependent — consistently biased toward input from the recent past — which stabilizes our visual experience in the face of noisy input. Several characteristics of visual serial dependence suggest that it may be tuned to the dynamics of the natural world, operating selectively across inputs that likely arose from the same physical object. For example, stimuli that move a large distance between trials are released from serial dependence, consistent with the fact that real-world objects do not spontaneously jump to entirely new locations. Real-world objects also do not undergo sudden large rotations on the retinae; rather, they undergo small, smooth rotations, often due to head and body movements. Is visual serial dependence tuned to object rotation, occurring most strongly across stimuli with small rotations, as expected of natural visual input? Here, we tested the rotational tuning of serial dependence using a shape judgment task. Observers saw a sequence of shapes drawn randomly from a morphed shape continuum, each presented with a random rotation (500ms presentation, one stimulus every ~5 seconds). On each trial, observers reported the shape they just saw using an adjustment response. We found that the perceived shape on a given trial was attracted toward the shape seen on the previous trial (serial dependence). The strength of this attraction did not fall off monotonically with increasing stimulus rotation between trials as we had predicted. Rather, the serial dependence from one trial to the next was governed by the pixel-wise image similarity of successive stimuli. Our results suggest that serial dependence is tolerant to object rotations, in contrast to its systematic tuning to position changes. In the face of object rotation, low-level image similarity is a key determinant of the strength of serial dependence across successive inputs.
Meeting abstract presented at VSS 2016