Diffusion-concentration neural system for separating figures from background

Abstract

According to psychology, diffusion and concentration of neural activity is a basic law of the neuronal interaction in the cerebral neocortex. There exists evidence that the lateral connections between cells in the cerebral cortex take the form that the short-range lateral connections are excitatory and the long-range ones are inhibitory [12]. This paper shows that the positive short-range connections perform the diffusional function, while the negative long-range ones perform the concentrative function. Diffusion and concentration are two reverse and coexistent neural processes whose relative strength can be controlled by these short-range or long-range connections. When the diffusion is strong relative to the concentration, all neurons in the whole cortex keep low and homogeneous active levels; when the concentration is stronger, the active level of these around the focus of the concentration will rise sharply. The diffusion and concentration are perhaps an essential neural mechanism that occurs at different stages of vision. This paper uses the mechanism to separate figures from background and presents a neural structure, called the diffusion-concentration network (DCN). DCN is a 2-D array of neurons with both the positive variable short-range and negative constant long-range lateral connections. Attention which serves as the top-down input of DCN is the diffusional signal, that is, the diffusional source, and spreads over the network. Edges information which serves as the bottom-up input is the blocking signal and inhibits the positive short-range connections to block the contour-sensitive diffusive process. As a result, in the positions across edges the diffusion becomes weak and the concentration becomes dominant. The diffusion-blocking process has the active level of the neurons within a contour rise, and further it will give the directions of concentration and instruct the active potential of neurons in transfer from the neurons in the background region to those in the figure regions. The concentration embodies the psychophysical result that figures and background are in competition with each other. The competition makes both figures strengthened and background suppressed. The computer simulation of the network is given. The difference from previous approaches is discussed.