Figure 16From: A neural computational model for bottom-up attention with invariant and overcomplete representationInfluences of different parameters in our model with invariant features on an illusory contour. (a) source image; (b) result of optimized parameters; (c) smaller RF size in layer 1, which still detects the s contour but in a less continuous way (discrete points can be seen on the s contour); (d) bigger RF size in layer 1, which produces a more continuous s curve and no obvious border effects; (e) smaller RF size in layer 2, which does not influence the result significantly (discrete points can be seen on the s contour); (f) smaller coefficient αwhich produces less strong suppression; (g) smaller excitation bandwidth σ ex , which detects a weak s contour.Back to article page