Burgess A E
Center for Imaging Science, Rochester Institute of Technology, New York 14623-5604.
J Opt Soc Am A Opt Image Sci Vis. 1994 Apr;11(4):1237-42. doi: 10.1364/josaa.11.001237.
Research on human-observer performance for noise-limited tasks (such as those found in medical imaging) has recently progressed to investigations in which some signal or image parameters are statistically defined. In these cases the ideal-observer procedure is usually nonlinear, and analysis is mathematically intractable. Two suboptimal but linear observer models have been proposed for mathematical convenience. The Hotelling observer is the optimal linear model and has been found to give a good fit to most human results. The nonprewhitening (NPW) matched filter also has been useful for explanation of some human results. Rolland and Barrett [J. Opt. Soc. Am. A 9, 649 (1992)] recently reported human results for detection of signals in white noise superimposed on statistically defined (lumpy) backgrounds in experiments that simulated nuclear medicine imaging systems. They found that the Hotelling model gave a good fit, whereas the simple NPW matched filter gave a poor fit. It is shown that the NPW model can also fit their data if a spatial frequency filter of a shape similar to the human contrast-sensitivity function is added to the NPW observer model. The best fit is achieved by use of an eye-filter model E(f) = f1.3 exp(-cf2), with c selected to yield a peak at 4 cycles/deg.
针对噪声受限任务（如医学成像中的任务）的人类观察者性能研究，近来已发展到对一些信号或图像参数进行统计定义的研究。在这些情况下，理想观察者程序通常是非线性的，且分析在数学上难以处理。为了数学上的便利，已提出了两种次优但线性的观察者模型。霍特林观察者是最优线性模型，已发现它能很好地拟合大多数人类实验结果。非白化（NPW）匹配滤波器对解释一些人类实验结果也很有用。罗兰和巴雷特[《美国光学学会志A》9, 649 (1992)]最近报告了在模拟核医学成像系统的实验中，人类对叠加在统计定义（块状）背景上的白噪声中的信号进行检测的结果。他们发现霍特林模型拟合得很好，而简单的NPW匹配滤波器拟合得很差。结果表明，如果在NPW观察者模型中加入一个形状类似于人类对比敏感度函数的空间频率滤波器，NPW模型也能拟合他们的数据。通过使用眼滤波器模型E(f) = f1.3 exp(-cf2)可实现最佳拟合，其中c的选择要使得在4周/度处出现一个峰值。
