engleski

Boundary Computation, Presynaptic Inhibition, and Lightness Perception

Munker-White illusion, Benary's cross, checkerboard contrast and second-order simultaneous contrast are examples of perceptual phenomena that could not be explained by classical concepts such as Wallach's ratio rule and center-surround antagonism. Recent theoretical contributions emphasize importance of depth separation (Anderson, 1997 ; Grossberg, 1997). For instance, darkening of gray patches on white stripes in Munker-White illusion is viewed as a consequence of separating gray patches from black occluder on the same depth plane with white background followed by simultaneous contrast. However, such analyses do not preclude possibility that monocular mechanisms may contribute in formation of these illusions. Since all illusions mentioned involve two or more aligned borders with different magnitude of contrast it is proposed that low contrast contour receives stronger support if it is aligned with high contrast contour and therefore performs contrast negation in filling-in layer (Ross & Pessoa, 1995). In order to implement this proposition bipole cell model is revised. New model exhibits analog sensitivity and operates as a statistical MAX gate rather than AND gate. Both features are consequences of presynaptic inhibition embedded in cooperative interactions. MAX gate means that bipole cell inherits activity level from the lobe that samples stronger contrast. Computer simulations are performed with extension of neural network for lightness perception formulated in the tradition of filling-in theories (Ross & Pessoa, 1995). The network has four stages: 1) cells with center-surround receptive fields ; 2) simple and complex cells ; 3) bipole cells ; and 4) filling-in which combines signals from the first and fourth stage. Activity distributions in filling-in layer show that model correctly predicts appearance of gray patches in all mentioned illusions. By observing that low contrast stimuli disable contrast negation, model could account for variants of Munker-White illusion presented by Anderson (1997). Model behavior is consistent with relative luminance relationships necessary for Munker-White illusion to appear (Spehar et al., 1995). Moreover, model could be easily extended in order to handle background constancy, crispening effect and stimuli investigated by Agostini & Proffitt (1991).

neural networks; vision; presynaptic inhibition; lightness illusions

nije evidentirano