Which way is down? Positional distortion in the tilt illusion

Abstract

Contextual information can have a huge impact on our sensory experience. The tilt illusion is a classic example of contextual influence exerted by an oriented surround on a target's perceived orientation. Traditionally, the tilt illusion has been described as the outcome of inhibition between cortical neurons with adjacent receptive fields and a similar preference for orientation. An alternative explanation is that tilted contexts could produce a re-calibration of the subjective frame of reference. Although the distinction is subtle, only the latter model makes clear predictions for unoriented stimuli. In the present study, we tested one such prediction by asking four naive subjects to estimate three positions (4, 6, and 8 o'clock) on an imaginary clock face within a tilted surround. To indicate their estimates, they used either an unoriented dot or a line segment, with one endpoint at fixation in the middle of the surround. The surround's tilt was randomly chosen from a set of orientations (± 75°, ± 65°, ± 55°, ± 45°, ± 35°, ± 25°, ± 15°, ± 5° with respect to vertical) across trials. Our results showed systematic biases consistent with the tilt illusion in both conditions. Biases were largest when observers attempted to estimate the 4 and 8 o'clock positions, but there was no significant difference between data gathered with the dot and data gathered with the line segment. A control experiment confirmed that biases were better accounted for by a local coordinate shift than to torsional eye movements induced by the tilted context. This finding supports the idea that tilted contexts distort perceived positions as well as perceived orientations and cannot be readily explained by lateral interactions between orientation selective cells in V1.

Figure 1. Examples of contextual effects in vision.

Ebbinghaus illusion: although the two orange circles are exactly the same size, the one on the left appears smaller by virtue of the size of the surrounding circles (Ebbinghaus, 1897). b) Mach bands: illusory dark or light stripes are perceived next to the boundary between two regions of an image with different lightness gradients (Mach, 1865). c) Brightness contrast: the left end of the horizontal bar appears to be brighter than the right one, depending on the brightness of the surround. In fact, the bar is just one color (Heiring, 1878). d) Contrast contrast: a low contrast texture surrounded by an uniform background seems to have higher contrast than the same one but surrounded by a high-contrast texture (Chubb et al., 1989).

Figure 2. Angular function of the tilt illusion.

The plot shows the bias magnitude as a function of the angle difference between surround and target orientations. When a vertically oriented grating is surrounded by a context tilted 15° away (top inset), the visual system exaggerates the difference between their orientations giving rise to the phenomenological repulsion of the vertical stimulus from the surround orientation. For surround-centre angles larger than 60° the illusion is inverted so that the vertical stimulus appears attracted toward to the surround's orientation.

Figure 3. Labels and tuning shift models of contextual influence.

Left: tuning functions of cortical orientation detectors. When a vertical target is presented alone the vertically selective neurons (red curve) responds strongly (solid circle) giving rise to the perception of a vertical stimulus (red circled perceptual label). Upper row: label shift model. Perceptual labels (oriented Gabor patches) shift towards units aligned with the visual context so that vertical orientations are perceived as tilted away. Lower row: tuning shift model. Tuning curves are attracted by the surround orientation causing a re-calibration of the vertical towards the surround orientation. Since the perceptual label stay put, the vertical stimulus will excite units labelled as tilted away from vertical, giving rise to a repulsive illusion.

Figure 4. Example stimuli and procedure for the main experiment.

a–b) Stimuli. On each trial the surround's orientation was randomly drawn from the set θ ∈ {±75, ±65, ±55, ±45, ±35, ±25, ±15, ±5}, and the phase was randomised. In the example shown above, the surrounds are tilted 25° clockwise with respect to vertical. On average, our observers showed systematic biases consistent with the tilt illusion in both the conditions. c) Cartoon of the task. Observers were requested to align the pointer to one of three possible positions (indicated in red in the figure) on an imaginary clock face by pressing either the left or right arrow key. The trial was terminated by the observer pressing the space bar.

Figure 5. The effect of pointer type and position on alignment biases.

Average unsigned alignment biases segregated on the basis of pointer and position. Positive biases with the segment are consistent with the direct tilt illusion. Positive biases with the point are in the same direction. Estimates of the 6 o'clock position were both more precise and more accurate than estimates of the 4 and 8 o'clock positions. Biases were just as large (if not even larger) when observers indicated the target position with an isotropic dot. In all plots error bars contain 2 SEs.

Figure 6. The effect of pointer type and position: individual data.

Average alignment biases for each subject, segregated on the basis of pointer and position. Plots show the data from four subjects (2 rows for subject) Also the individual data show an higher precision for estimates of the 6 o'clock position. Biases for dot and segments look pretty much similar. In all plots error bars indicate 95% confidence intervals.

Figure 7. Example stimuli and procedure for the control experiment.

a) Stimuli. Two sinusoidal grating annuli were flashed for 100 ms on the right and left side of a central fixation point. The angle of the dots positioned inside each annulus varied symmetrically among trials in accordance with a staircase adaptive algorithm. b) Cartoon of the task. Observers had to press the left or right arrow keys to report whether the two dots' positions appeared inward or outward with respect to the central fixation point.

Figure 8. The effect of contextual orientation on the induced position bias.

Point of subjective verticality estimated for each annulus' tilt and collapsed over observers. This configuration produces even larger biases, consistent with previous studies of the tilt illusion. In all plots error bars contains 2 MSE.

Figure 9. The effect of contextual orientation on position bias, individual data.

Point of subjective verticality estimated for each annulus' tilt for each observer. Aside from intersubjective differences in magnitude, the pattern of position biases is fairly consistent across subjects. In all plots error bars contains 95% confidence intervals.