Construction of Neural Network acquiring an Internal Model of Visual Perceptual Grouping

点击翻译按钮获取中文摘要

In the human visual system, the visual modalities are detected at visual cells in the retina, the lateral geniculate nucleus(LGN) and the primary visual cortex (V1). Aperture problem is a kind of experiments for analyzing the binding mechanism for motion processing in the early visual system. A circle aperture where a bar is moving in the background is first displayed at the computer display. Two other circles appear next at both sides of the center circle, but two bars are also moving in the background, and the bar's orientation is different from the center's. If subjects perceived three bars as a bar, the center bar's moving orientation would be changed as same as at the both sides of circle's. This perceptual grouping is strongly depending on the display time. On the other hand, visual models and neural networks based on the human visual system have been proposed, e.g., BCS, FCS, ARTMAP, fuzzy ARTMAP and TAM Network. TAM (Topographic Attentive Mapping) Network is a biologically-motivated neural network. TAM Network is analogous to receptive field, LGN, and V1 in the structure from the input layer to the output layer. When the network makes an incorrect error, the attentional mechanism is invoked based on the feedback signals and inhibitory synapses, and the error is adjusted to be smaller. In this research, the following researches are studied.1.Perceptual rates are estimated changing the display time, radius, distance between circles, and the dependency of display time on radius and distance between circles is confirmed.2.The curve of perceptual rates for display time is convex.3.Fuzzy rules are acquired from aperture's data using TAM Network, and the usefulness of feedback signals and inhibitory synapses of TAM Network is shown.By these researches, we show the possibility of the decreasing of perceptual rate in the late display time more than 550ms. The usefulness of the feedback signals and the inhibitory synapses of TAM Network is also shown.