Yan Shiju, Wang Chengtao, Ye Ming
Institute of Biomedical Manufacturing and Life Quality Engineering, School of Mechanical Engineering, Shanghai Jiaotong University, Shanghai, China.
Med Phys. 2007 Nov;34(11):4194-206. doi: 10.1118/1.2791037.
This paper presents a novel integrated method to correct geometric distortions of XRII (x-ray image intensifier) images. The method has been compared, in terms of mean-squared residual error measured at control and intermediate points, with two traditional local methods and a traditional global methods. The proposed method is based on the methods of moving least squares (MLS) and polynomial fitting. Extensive experiments were performed on simulated and real XRII images. In simulation, the effect of pincushion distortion, sigmoidal distortion, local distortion, noise, and the number of control points was tested. The traditional local methods were sensitive to pincushion and sigmoidal distortion. The traditional global method was only sensitive to sigmoidal distortion. The proposed method was found neither sensitive to pincushion distortion nor sensitive to sigmoidal distortion. The sensitivity of the proposed method to local distortion was lower than or comparable with that of the traditional global method. The sensitivity of the proposed method to noise was higher than that of all three traditional methods. Nevertheless, provided the standard deviation of noise was not greater than 0.1 pixels, accuracy of the proposed method is still higher than the traditional methods. The sensitivity of the proposed method to the number of control points was greatly lower than that of the traditional methods. Provided that a proper cutoff radius is chosen, accuracy of the proposed method is higher than that of the traditional methods. Experiments on real images, carried out by using a 9 in. XRII, showed that residual error of the proposed method (0.2544 +/- 0.2479 pixels) is lower than that of the traditional global method (0.4223 +/- 0.3879 pixels) and local methods (0.4555 +/- 0.3518 pixels and 0.3696 +/- 0.4019 pixels, respectively).
本文提出了一种用于校正X射线图像增强器（XRII）图像几何失真的新型集成方法。已根据在控制点和中间点测量的均方残差误差，将该方法与两种传统局部方法和一种传统全局方法进行了比较。所提出的方法基于移动最小二乘法（MLS）和多项式拟合。对模拟和真实的XRII图像进行了大量实验。在模拟中，测试了枕形失真、S形失真、局部失真、噪声以及控制点数量的影响。传统局部方法对枕形和S形失真敏感。传统全局方法仅对S形失真敏感。发现所提出的方法既对枕形失真不敏感，也对S形失真不敏感。所提出的方法对局部失真的敏感度低于传统全局方法或与之相当。所提出的方法对噪声的敏感度高于所有三种传统方法。然而，只要噪声的标准差不大于0.1像素，所提出方法的精度仍高于传统方法。所提出的方法对控制点数量的敏感度远低于传统方法。只要选择合适的截止半径，所提出方法的精度就高于传统方法。使用9英寸XRII对真实图像进行的实验表明，所提出方法的残差误差（0.2544±0.2479像素）低于传统全局方法（0.4223±0.3879像素）和局部方法（分别为0.4555±0.3518像素和0.3696±0.4019像素）。
