Abstract
Background: According to the optimal object tracking model (Kwon, Tadin, & Knill, 2015, PNAS), the visual system integrates noisy sensory inputs with a forward model to estimate positions and motions of moving objects. The model provides a unifying account of a wide range of visual phenomena related to the integration of motion and position signals. However, illusory perception of flicker-defined motion (Mulligan & Stevenson, 2014, VSS) provides a counterexample. The flicker-defined motion appears to jump when it moves continuously, which conflicts with the prediction of the optimal tracking model.
Model: The propagation noise of a tracking model represents the system's assumption of the random changes of velocity, and conventionally it was assumed to follow a Gaussian distribution. Considering that various natural movements follow a fat-tailed distribution (Kleinberg, 2000, Nature), we built a model that assumes a fat-tailed propagation noise and found that the model can predict jumping percepts of the flicker-defined motion.
Experiment: We asked participants to report the perceived jumping frequency across five object speeds (3.7, 5.5, 9.2, 11°/s), two pattern speeds (1X object speed, 2X object speed) and two eccentricities (11, 16°) by adjusting the jumping frequency of a probe stimulus. Results show that the jumping frequency increases as (a) the object speed increases, (b) the relative pattern speed increases, or (c) the eccentricity decreases. Results are consistent with the prediction of the fat-tailed model and rule out the possibility that the observed jumping percept is due to periodic attentional sampling. Moreover, the same model can account for a range of other motion phenomena.
Conclusion: The fat-tailed propagation noise model can account for a broader range of perceptual phenomena than the model based on commonly used Gaussian noise. Evidently, the visual system assumes fat-tailed propagation noise, noise that closely mirrors statistics of object movements observed in nature.