Wilson H R, Wilkinson F
Visual Sciences Center, University of Chicago, IL 60637, USA.
Vision Res. 1998 Oct;38(19):2933-47. doi: 10.1016/s0042-6989(98)00109-6.
Glass (Nature 1969;223:578-580) patterns are random dot stimuli that generate a percept of global structure. To study the mechanisms underlying this global form perception, concentric, radial, hyperbolic, and parallel Glass patterns were constructed. Thresholds for detecting each type of pattern were measured by degrading the patterns through the addition of noise. Concentric patterns yielded the lowest thresholds for all subjects, while radial and hyperbolic patterns produced somewhat higher thresholds. For all subjects the parallel patterns produced the highest thresholds. Threshold measurements as a function of the area containing pattern structure provided evidence for global pooling of orientation information in the detection of radial and concentric Glass patterns but only local pooling in the detection of parallel patterns. Monte-Carlo simulations demonstrate that plausible neural models can accurately predict the data. These models indicate that the visual system contains networks that pool orientation information within regions 3.5-4.5 degrees in diameter in central vision. This pooling is organized to extract cross-shaped, X-shaped, and quasi-circular forms from the retinal image. The results are in good agreement with recent single unit physiology of primate area V4, an intermediate level of the form vision pathway.
格拉斯（《自然》，1969年；223卷：578 - 580页）图案是能产生整体结构感知的随机点刺激。为研究这种整体形状感知背后的机制，构建了同心、放射状、双曲线状和平行的格拉斯图案。通过添加噪声使图案退化来测量检测每种图案的阈值。对于所有受试者，同心图案的阈值最低，而放射状和双曲线状图案的阈值略高。对于所有受试者，平行图案的阈值最高。作为包含图案结构区域函数的阈值测量为在检测放射状和同心格拉斯图案时方向信息的全局合并提供了证据，但在检测平行图案时仅为局部合并。蒙特卡罗模拟表明，合理的神经模型能够准确预测这些数据。这些模型表明，视觉系统包含在中央视觉中对直径3.5 - 4.5度区域内的方向信息进行合并的网络。这种合并被组织起来从视网膜图像中提取十字形、X形和准圆形形式。这些结果与灵长类动物V4区（形状视觉通路的一个中间层次）最近的单神经元生理学研究结果高度一致。
