Appendix 2—figure 7. Projected image of Sin(x² + y²) function is used to illustrate the effect of aliasing and how stochastic variability in the sampling matrix combats aliasing effectively.

(A) Sin(x²+y²) is plotted with 0.1 resolution. (B) Under-sampling the same image by an ordered matrix leads to aliasing: ghost rings appear when the image (of the function) is sampled with 0.2 resolution. Aliasing is a critical problem, as the nervous system cannot differentiate the fake image rings from the original real image. (C) Sampling the image (A) with a random matrix may lose some of its fine resolution, due to broadband noise, but such sampling is anti-aliasing; sampling with random points at 0.2 resolution. (D) Color photoreceptor distributions across macaque (Field et al., 2010) (red, green and blue cones; left) and Drosophila retina (Vasiliauskas et al., 2011) (R7y and R7p receptors; right) show random-like sampling matrixes, suggesting that this sampling matrix sensitivity randomization would have an anti-aliasing role. (E) Crucially, by integrating and redistributing R1-R6 outputs with additional gap-junctional inputs from randomized R7/R8 color channels (Wardill et al., 2012) (D) and Appendix 2—figure 6) for each image pixel during synaptic transmission to LMCs, any broadband sampling noise should be much reduced and the R1-R6 (motion) channel’s spectral range whitened (Wardill et al., 2012). Note how LMC output peaks before the corresponding R1-R6 output. Scale bars: 10 mV / 20 ms. Sub-figure (E) is modified from (Wardill et al., 2012).