A Bayesian account of context-induced orientation illusions

1
ECVP 2012, 2-6 September, Alghero, Italy
A Bayesian account of context-induced orientation
illusions Endel PõderInstitute of Psychology,
University of Tartu, Estonia E-mail
endel.poder_at_ut.ee
Background In classical tilt illusion (Gibson,
1937, Goddard et al., 2008), perceived
orientation of a target object is repulsed from
the orientation of surrounding context. My
earlier study (Põder 2009, 2012) with crowded
orientation stimuli showed strong assimilation
bias at about the same target-surround
differences, opposite to the classical tilt
illusion.
Model I use a Bayesian model with conditional
priors (e.g. Stocker Simoncelli, 2008). I
suppose that observer automatically categorizes
stimulus patterns as either homogeneous or target
pop-out. Probability of pop-out increases with
increasing target-flanker difference. The target
likelihood function is combined with the
different prior distributions for the pop-out and
homogeneous trials. Responses are calculated as
the means of posterior distributions. Total
distribution of responses is a mixture of the two
distributions with opposite biases.
Perceived as homogeneous
Target pop-out
Purpose To understand the mechanisms of
contextual effects in orientation perception.
Experiment Contextual effects were measured with
the same stimuli in fovea and periphery. In
fovea, visibility was varied by using different
contrasts.

• Methods
• Stimuli A target Gabor (3.2 cpd) surrounded by 6
  flanking Gabors, target-flanker distance 0.8 deg,
  target-flanker orientation difference varied from
  0 to 64 deg.
• Exposure duration 24 ms.
• Task to reproduce the target orientation by
  rotating a probe Gabor.
• Three conditions
• Central presentation, high contrast (0.92)
• Central presentation, low contrast (0.17)
• Peripheral presentation (eccentricity 5 deg,
  contrast 0.92).

Results of experiment
Results of simulation
Conclusions The experimental results suggest that
perceived homogeneity vs. target pop-out may
determine the nature of illusion. A Bayesian
model with context-dependent priors can produce
qualitatively similar results.
References Gibson, J. J. (1937). Adaptation,
after-effect, and contrast in the perception of
tilted lines II. Simultaneous contrast and the
areal restriction of the after-effect. Journal of
Experimental Psychology, 20, 553569. Goddard,
E., Clifford, C. W., Solomon, S. G. (2008).
Centre-surround effects on perceived orientation
in complex images. Vision Research, 48,
13741382. Põder, E. (2009). Integration of
crowded orientation signals. Perception, 38, ECVP
Supplement, 179. Põder, E. (2012). On the rules
of integration of crowded orientation signals.
i-Perception, 3, 440454. Stocker, A. A.,
Simoncelli, E. P. (2008). A Bayesian model of
conditioned perception. In J. C. Platt, D.
Koller, Y. Singer, S. Roweis (Eds.), Advances
in Neural Information Processing Systems. (pp.
14091416). Cambridge, MA MIT Press.
Reduced visibility (higher internal noise) of the
target orientation increases the range of
target-surround differences where assimilation
bias is observed.