Figure 1.
The dotted curve is the motion generated by the computer, and the solid curve is the motion reported by the observer.
χ is the maximum value of the normalized cross correlation function and τ is the time delay at which χ occurs.
Table 1.
Default values and range of various parameters used in experiments.
Figure 2.
(a) Block schematic of the model (b) Optical flow output by Watson Ahumada motion detector (c) Model response at various other stages in the pipeline.
Table 2.
Default values of model parameters used in simulations.
Figure 3.
Response reproducibility ζ vs. c.
Both model and humans show zero reproducibility at c = 0, and the reproducibility steadily increases with c, because the motion signal gets stronger.
Figure 4.
(a) histogram of Inter Flip Interval (IFI) at c = 0 (mode≈2 s) (b) normalised histogram of ln(IFI) together with a Gaussian fit.
The pdf of ln(IFI) given by the model is also shown.
Figure 5.
(a) χ vs. c (b) τ vs. c (fd = 30 ms, ic = 7°, dd = 5 dots/degrees2) (c) τ vs. χ scatter plot and piecewise linear fit for experimental data.
Figure 6.
c = 0.1, ic = 7°, dd = 5 dots/degrees2.
Figure 7.
c = 0.2, ic = 7°, fd = 30 ms. Model simulations done at 256×256 pixel resolution for dd>10.
Table 3.
Probability of mismatch values for dot densities in Figure 7.
Figure 8.
(a) χ vs. hop size h for human observers (b) χ vs. hop size for model.
c = 0.4, fd = 30 ms, ic = 7°.
Figure 9.
χ vs. angle subtended by inner circle diameter ic.
Angle subtended by outer circle diameter is fixed at 10°. c = 0.1, dd = 5 dots/degrees2, fd = 30 ms. Model simulations at 256×256 pixel resolution. f0 denotes center frequency of Watson Ahumada sensors.
Figure 10.
χ vs. c for contrast reversing dots.
fd = 30 ms, ic = 7°, dd = 2.5 dots/degrees2.
Figure 11.
The Fourier Transform of an image undergoing coherent translational motion + periodic reverse contrast lies on infinitely many planes of the form , with
being an odd number.
The dashed lines denote the window of visibility [30].
Figure 12.
Point O represents the true center of rotation, whereas point C is the center relative to which rotary motion is computed by the model.
The offset is given by where
is radius of inner circle.
Figure 13.
χ vs. center relative to which rotary motion is computed (a) full 360° annulus is visible (b) only 90° of annulus is visible.
Type1 – a single 90° sector is visible. Type 2 - two sectors located diametrically opposite to each other, and each 45° in size, are visible. Both curves are for the model.
Figure 14.
In case of type 1 only one sector is displayed, whereas in case of type 2 two sectors located diametrically opposite to each other, and each half the size of sector in type 1, are displayed. (a) human performance (b) model performance.
Figure 15.
Effect of inserting K random frames between correlated frames.
c = 0.5, fd = 30 ms, dd = 5 dots/degrees2, ic = 7°.
Figure 16.
(a) tangential dipoles (b) radial dipoles.
Center-to-center spacing = 17′ in both cases.
Figure 17.
χ vs. bwir (black to white intensity ratio).
Dipole spacing = 6′, c = 0.5, dd = 2.5 dots/degrees2, RC ON, (a) human observers (b) Watson Ahumada model with simulations at 256×256 pixel resolution.