A software tool for measurement of the modulation transfer function
A software tool for measurement of the modulation transfer function
复制标题
用于测量调制传递函数的软件工具
DOI：
--
复制
发表时间：
2005
期刊：
影响因子：
--
通讯作者：
Andrew D. A. Maidment
中科院分区：
文献类型：
--
作者：
Y. Kao;M. Albert;A. Carton;H. Bosmans;Andrew D. A. Maidment
调制传递函数（MTF）的计算是一个多步骤的过程。在计算的每个步骤中，算法可能具有与成像系统或物理学无关的固有误差。我们设计了一个具有图形用户界面的软件工具，以便于计算MTF和分析这些计算的准确性。为了最小化误差源，使用没有任何噪声或伪影的模拟边缘图像。我们首先检查了常用的边缘斜率估计算法的准确性，即逐行微分，然后进行线性回归拟合。论证了边缘长度和边缘相位对线性回归算法的影响。同时，给出了边缘斜率估计误差与调制传递函数误差之间的关系。我们比较了两种核[-1，1]和[-1，0，1]在从边缘扩展函数（ESF）的有限元微分计算线扩展函数（LSF）中的性能。我们发现选择IEC推荐的[-1，0，1]内核没有实际优势。然而，应应用有限元微分的校正;否则，MTF中存在可测量的误差。最后，我们将噪声添加到边缘图像中，并比较了ESF上的两种降噪方法的性能：具有boxcar核的卷积和单调性约束。前一种方法在采样频率范围内的MTF误差始终大于4%，而后一种方法的误差始终小于1%。
Calculation of the modulation transfer function (MTF) is a multi-step procedure. At each step in the calculation, the algorithms can have intrinsic errors which are independent of the imaging system or physics. We designed a software tool with a graphical user interface to facilitate calculation of MTF and the analysis of accuracy in those calculations. To minimize the source of errors, simulated edge images without any noise or artifacts were used. We first examined the accuracy of a commonly used edge-slope estimation algorithm; namely line-by-line differentiation followed by a linear regression fit. The influence of edge length and edge phase on the linear regression algorithm is demonstrated. Furthermore, the relationship of edge-slope estimation error and MTF error are illustrated. We compared the performance of two kernels, [-1,1] and [-1,0,1], in the computation of the line spread function (LSF) from finite element differentiation of the edge spread function (ESF). We found that there is no practical advantage in choosing the [-1,0,1] kernel, as recommended by IEC. However, a correction for finite element differentiation should be applied; otherwise, there is a measurable error in the MTF. Finally, we added noise into the edge images and compared the performance of two noise reduction methods on the ESF; convolution with a boxcar kernel and a monotonicity constraint. The former method always produces MTF error higher than 4% up to the sampling frequency, while the latter was consistently less than 1%.
影响因子：
3.8
作者：
Maidment, ADA;Albert, M
通讯作者：
Albert, M