Beckmann Paul J, Legge Gordon E
Department of Psychology, University of Minnesota, 75 East River Road, Minneapolis, Minnesota, 55455, USA.
J Opt Soc Am A Opt Image Sci Vis. 2002 Dec;19(12):2349-62. doi: 10.1364/josaa.19.002349.
We created a sequential ideal-observer model that could address the question, How much of letter identification performance and its change with eccentricity can be accounted for by preneural factors? The ideal-observer model takes into account preneural factors including the stimulus rendering properties of a CRT display, the optical imaging quality of the eye, and photon capture and sampling characteristics of the cones. We validated the formulation of the model by comparing its performance on simple psychophysical tasks with that of previous sequential ideal-observer models. The model was used to study properties of the image rendering of letters. For example, the model's identification of high-resolution letters (i.e., many pixels per letter), but not low-resolution letters, is largely immune to changes in pixel width. We compared human and ideal-observer letter-identification acuity for the lowercase alphabet at 0 degrees, 5 degrees, and 20 retinal eccentricity. Acuity of the ideal observer for high-contrast letters is approximately seven times better than that of the human observers at 0 degrees. Acuity decreased with eccentricity more rapidly for human observers than for the ideal observer such that the thresholds differed by a factor of 50 at 20 degrees. A decrease in stimulus duration from 100 to 33 ms resulted in no decrease in relative threshold size between the human and ideal observers at all eccentricities, indicating that humans effectively integrate stimulus information over this range. Decreasing contrast from 75% to 25%, however, reduced the difference in acuities twofold at all eccentricities between humans and the ideal-observer model, consistent with the presence a compressive nonlinearity only in the human observers. The gap between human and ideal acuity in central vision means that there are substantial limitations in human letter recognition beyond the stage of photoreceptor sampling. The increasing performance gap between human and ideal-observer performance with eccentricity implicates an increasing role of neural limitations with eccentricity in limiting human letter identification.
我们创建了一个序贯理想观察者模型，该模型能够回答这样一个问题：神经前因素能够在多大程度上解释字母识别性能及其随偏心率的变化？理想观察者模型考虑了神经前因素，包括阴极射线管显示器的刺激呈现特性、眼睛的光学成像质量以及视锥细胞的光子捕获和采样特性。我们通过将其在简单心理物理学任务中的表现与先前的序贯理想观察者模型进行比较，验证了该模型的公式。该模型用于研究字母图像渲染的特性。例如，该模型对高分辨率字母（即每个字母有许多像素）的识别，而不是对低分辨率字母的识别，在很大程度上不受像素宽度变化的影响。我们比较了人类和理想观察者在0度（视网膜中心）、5度和20度视网膜偏心率下对小写字母表的字母识别敏锐度。对于高对比度字母，理想观察者在0度时的敏锐度大约比人类观察者高七倍。人类观察者的敏锐度随偏心率下降的速度比理想观察者更快，以至于在20度时阈值相差50倍。刺激持续时间从100毫秒减少到33毫秒，在所有偏心率下，人类和理想观察者之间的相对阈值大小均未降低，这表明人类能够在这个范围内有效地整合刺激信息。然而，将对比度从75%降低到25%，在所有偏心率下，人类和理想观察者模型之间的敏锐度差异减少了两倍，这与仅在人类观察者中存在压缩非线性一致。中央视觉中人类与理想敏锐度之间的差距意味着，在光感受器采样阶段之后，人类字母识别存在实质性限制。随着偏心率增加，人类与理想观察者性能之间的差距越来越大，这意味着偏心率相关的神经限制在限制人类字母识别方面的作用越来越大。
