亮度對比在知覺圖形處理之增益控制機制

Abstract

Glass圖形(Glass, 1969)是一群由兩點所組成的「雙點」隨機分佈組成，我們所知覺到的圖形取決於局部雙點的幾何位置排列。本實驗中我們使用一個由一群「三點」隨機分佈組成的新圖型，這些三點之間等距，這些三點中最接近圖形的中心的點稱之為「錨」(anchor)。如果藉由操弄三點各自的亮度對比，使得視覺系統在辨識整體圖形時將錨與其中一邊的點連成一對，則整體可看見一個方向的螺旋結構；反之若錨與另一個邊點連在一起，則另一方向旋轉的圖形可預見。在此同時，未被串聯第三點的亮度對比也同時影響我們對圖形的判斷。本研究旨在研究藉由操弄局部三點各自的亮度對比是如何影響到我們知覺系統是如何將其中兩點整合成雙點進而影響後續我們對整體圖形的辨識。實驗情境中錨的亮度對比分別選用-20分貝及-10.46分貝，而三點中另外兩邊點的亮度對比則各自隨機抽取使用-29.54到-0.92分貝之間的值。在強迫選擇派典(forced choice task)下，探究受試者會將這新圖形分類為正螺旋亦或是逆螺旋的機率。結果顯示隨著其中一邊的點的對比亮度提升時，與錨連成一對而形成整體圖形的機率也隨之上升，然而當這個邊點與錨的亮度差異過大時，機率又再度下降。進一步而言，邊點與錨連結在一起機率高峰並非是在兩點的對比亮度最相近處，而會受到未被串串聯第三點的調節而此第三點的亮度對比上升而往被串聯點為高亮度對比的方向移動。實驗數據後來以「對比增益控制」(contrast gain control)模型模擬我們知覺歷程，此調控歷程之輸出為一個別運算元的反應除以一群相關的運算單元的反應加總，而未被串聯第三點的影響則透過增加該群運算單元的量抑制整體的輸出作解釋。本研究結果支持人類視覺在處理亮度對比時需仰賴此增益控制機制的調節。

Glass patterns (Glass, 1969) consist of randomly distributed dot pairs and their orientation is determined by certain mathematical transformations. In this study, a new pattern with a group of randomly distributed triple dots (tripoles) was used. Each tripole comprises an “anchor dot,” which is close to the center of the pattern, and two side dots. The distance between the anchor dot and each side dot as well as between the two side dots is equal. By manipulating the luminance contrast of each dot in the tripoles, if in the tripoles, the visual system groups the anchor dot with one of the two side dots, and the Glass pattern will be perceived to spiral in one direction (and vice versa). Meanwhile, the other side dot, which is not being grouped, would systematically influences the process of the grouped dots. The purpose of this study is to see how the variation of luminance contrast of intra-tripoles affects the local grouping and, then, the global perception. The anchor-contrast was either -20 or -10.46dB, and the contrast of side dots ranged from -29.54 to -0.92dB. The study involved a forced-choice paradigm. The probability of an observer’s categorization of a tripole Glass pattern as counter-clockwise or clockwise pattern was measured. The result showed that as the contrast of a side dot increased, the probability of it being grouped with the anchor dot increased initially, and then, the probability dropped as the contrast of side dot is greatly different from the anchor dot. Furthermore, the peak of response curve was not at where the grouped dots are of similar contrast, but where they are modulated by the contrast of the dot, which was not being grouped. As the contrast of the side dot, which was not being grouped to the anchor dot, increased, the peaks moved toward where the grouped side dots were of high contrast. The result could be fitted with a divisive inhibition model, which was the response of each mechanism resulting from an output of the linear operator divided by the sum of relevant channels. Meanwhile, the shift of the response curves could be accounted for by the inhibition input, which suggests that the contrast processing of Glass pattern is mediated by a contrast gain-control mechanism.