A computational model of perceptual grouping and 3D surface completion in the mime effect

We propose a computational model of perceptual grouping for explaining the 3D shape representation of an illusory percept called “mime effect.” This effect is associated with the generation of an illusory, volumetric perception that can be induced by particular distributions of inducing stimuli such as cones, whose orientations affect the stability of illusory perception. The authors have attempted to explain the characteristics of the shape representation of the mime effect using a neural network model that consists of four types of cells–encoding (E), normalizing (N), energetic (EN), and geometric (G) cells. E cells represent both the positions and orientations of inducing stimuli and the mime-effect shape, and N cells regulate the activity of E cells. The interactions of E cells generate dynamics whose mode indicates the stability of illusory perception; a stable dynamics mode indicates a stable perception, whereas a chaotic dynamics mode indicates an unstable perception. EN cells compute the Liapunov energetic exponent (LEE) from an energy function of the system of E cells. The stable and chaotic dynamics modes are identified by strictly negative and strictly positive values of LEE, respectively. In case of stability, G cells perform a particular surface interpolation for completing the mime effect shape. The authors confirm the model behaviour by means of computer-simulated experiments. The relation between the model behaviour and the shape representation in the human brain is also discussed.