摘要
CT技术不仅可以精确地、清晰地显示物体内部细节的结构关系及缺陷状况,还可以定量地给出细节的辐射吸收数据及物质组成,已在医学的病情诊断和工业的缺陷识别及属性测量中得到广泛的应用。三代扫描模式具有扫描速度快的优点,是目前医学CT和工业CT采用的主要扫描模式。环形伪影作为三代扫描模式的一个重要问题,不仅影响了图像的质量,还给图像噪声处理以及图像分割等后续处理造成困扰,降低了图像的识别能力和测量精度。这个问题工业CT成像系统检测中更是突出,因此,环形伪影的去除是CT图像预校正的关键一步。
根据重建图像中的环形伪影在投影正弦图中表现为竖直方向直线的特征,从投影正弦图的角度,提出两种环形伪影校正算法:
第一种是基于投影曲线高通滤波的环形伪影校正法。首先,采用S-L滤波函数对原始投影数据进行滤波,增强伪影信息;其次,对滤波后的投影数据进行投影积分,并对投影曲线采用微分处理,进一步增大伪影数据与工件轮廓数据的差异;然后,按照高频采样插值次数对差分后的投影数据进行均值滤波,并根据粗大误差判断准则选择阈值,确定伪影的位置;最后,分析传统线性插值校正不彻底的原因,采用线性外推插值来对伪影处投影数据进行校正。
第二种是基于改进的Canny算法的环形伪影校正法。在高斯滤波前,首先对原始投影数据采用S-L滤波器进行滤波,增强伪影信息;采用二维高斯函数的一阶偏导数构造滤波器计算梯度,降低噪声的影响;设定角度阈值对梯度方向进行限制,排除斜角方向边缘的检测,实现竖直方向边缘点检测;采用梯度阈值和链长度阈值实现伪影边缘点的检测与连接;分析了各行投影数据插值校正的问题,提出分段B样条拟合法对投影数据进行校正,实现多个连续的环形伪影校正。
通过对实际CT图像进行校正实验,结果表明,这两种校正方法都能有效地消除CT图像的环形伪影,又很好保持了图像边缘及分辨率。其中,第一种校正算法实现简单,对单个分散的环形伪影校正效果好,但是对多个连续的环形伪影校正受限;第二种实现了多个连续的环形伪影的校正,但是对于最大连接长度小于链长度阈值的伪影无法检测。它们各有特点,互相补充,但也不能完全互相取代。
CT technology not only is able to show that structure relation and defect status of object interior details accurately and clearly, but also can quantitatively give the radiation density data and material composition. So it is widely applied in medical illness diagnoses and industrial defect recognition and property measurement. Because of the advantages of rapid scanning, the third generations scanning mode has become dominating in MCT and ICT fields. As one of the important problems of the third generations scanning mode, ring artifacts not only effect the image quality, but also hamper subsequent processing, such as image noise processing, image segmentation, and so on. Finally, image recognition ability and measurement accuracy is reduced. This problem is particularly outstanding in ICT imaging system. Therefore, ring artifact removing is the key step to pre-correction of CT images.
According to the characteristics that the ring artifacts of CT image show vertical directional lines in the sinogram of CT images, two kinds of ring artifact correction algorithms were proposed based on the sinogram.
The first ring artifact correction method was based on projection profile high-pass filter. Firstly, the S-L filter was adopted to filter original projection data and enhance artifact information. Secondly, the filtered projection data was processed by projection integral, and then the projection profile was processed by differential to further increase the difference between artifacts edge data and work piece contour data. Thirdly, the differenced projection data was mean filtered according to the high-frequency sampling interpolation times, and artifacts’location was automatically searched according to threshold selected by gross error criterion. Finally, through analyzing the reasons why the ring artifacts could not be corrected by traditional linear interpolation, the linear extrapolation interpolation was proposed to correction the projection data of ring artifacts.
The second ring artifact correction method was based on the improved Canny algorithm. Firstly, the S-L filter was adopted to filter original projection data and enhance artifact information before using Gaussian filter. Then the gradient was acquired by the filter, which composed by 2d first-order partial derivative to reduce the noise affection. Thirdly, the angle thresholds are established for limiting gradient direction, particularly excluding diagonal direction edge detection and achieve vertical direction edge detection. Fourthly, a gradient threshold and an angle threshold were used to detect and connect marginal point of the artifacts. Finally, through analyzing the problems that the traditional linear interpolation algorithm is employed to correct the projection data, the subsection B-spline fitting method is proposed to implement multiple continuous ring artifacts correction.
Based on the actual CT images calibration experiment, the result indicates that the two proposed correction algorithms is able to remove the ring artifacts effectively and keep image edges and resolution unchanged. On one hand, the first correction algorithm is simple, and effective to remove single dispersed ring artifacts. But it is limited for multiple continuous ring artifacts. On the other hand, the second correction algorithm is able to reduce the multiple continuous ring artifacts, yet not effective to detect the ring artifacts which the maximum connecting length is smaller than chain length threshold. The two proposed methods have different characteristics, complement each other in some aspects,but can not totally replace.
引文
[1]张朝宗,郭志平.工业CT技术和原理[M].北京:科学出版社, 2009.
[2]谢强.计算机断层成像技术[M].北京:科学出版社, 2005.
[3]张俊哲.无损检测技术及其应用[M].北京:科学出版社, 2010.
[4]王珏,黄苏红,蔡玉芳.工业CT材料密度测量方法研究[J].计算机工程与应用, 2010, 46(2): 203-205.
[5] Jiang Hesieh. Reconstruction bias resulting from weighted projection and iso-center misalignment[J]. SPIE, 1999, 3661: 442-449.
[6] G. R. Davis, J. C. Elliott. X-ray microtomography scanner using time-delay integration for elimination of ring artifacts in the reconstructed image[J]. Nuclear Instruments And Methods in Physics Research A, 1997, 394:157-162.
[7]李俊江,路宏年,李保磊. X射线图像增强器像元响应不一致性的分析及校正[J].光学技术, 2006, 32(5): 779-784.
[8]王苦愚,张定华,黄魁东,等.一种锥束CT中平板探测器输出图像校正方法[J].计算机辅助设计与图形学学报. 2009, 21(7): 954-961.
[9]梁丽红,路宏年.射线面阵探测器成像系统校正研究[J].光子学报, 2004, 33(10): 1277– 1280.
[10]夏雄军.辐射图像噪声及CT环形伪影消除[D].北京:清华大学, 2007.
[11]李俊江,胡少兴,李保磊,等. CT图像环状伪影校正方法[J].北京航空航天大学学报, 2007, 33(11): 1382-1398.
[12] Jan Sijbers, Andrei Postnov. Reduction of ring artifacts in high resolution micro-CT reconstructions[J]. Physics in Medicine and Biology, 2004, 49(14): 1-8.
[13] Yang Jun, Zhen Xin, Zhou Linghong, et al.. Geometric correction for cone-beam CT reconstruction and artifacts reduction[C]. 2nd International Conference on Bioinformatics and Biomedical Engineering, shanghai, P.R. China: IEEE, 2008: 2386-2389.
[14] Chen Yenwei, Duan Guifen. Independent component analysis based ring artifact reduction in cone-beam CT images[C]. IEEE International Conference on Image Processing, November 7-12. 2009: 4189-4192.
[15]王珏,黄苏红,蔡玉芳.工业CT图像环形伪影校正[J].光学精密工程, 2010, 18(5): 1226-1233.
[16] Kowalski G. Suppression of ring artifacts in CT fan-beam scanners[J]. IEEE Trans On Nucl Sci, 1978, NS 25(5):1111-1116.
[17] Sun Haining, Qiu Shaokun, Lou SHanshan, et al.. A correction method for nonlinear artifacts in CT imaging[C]. Proceedings of the 26th Annual International Conference of the IEEE EMBS, San Francisco, P.R.USA: IEEE, 2004, 26 II:1290–1293.
[18] A. Matani, K. Terakawa. Artifact reduction filtering method for CT images[J]. Proceedings of the Second Joint EMBSBMES Conference. HOUSIM, TX. USA October 23-26. 2002: 1035-1036.
[19] Carsten Raven. Numerical removal of ring artifacts in microtomography [J]. Review of Scientific Instruments, 1998, 69(8):2978-2980.
[20] Beat Münch, Pavel Trtik, Federica Marone, et al.. Stripe and ring artifact removal with combined wavelet-Fourier filtering [J]. Optics Express, 2009, 17(10):8567-8591.
[21] Rainer Raupach. Method for removing rings and partial rings in computed tomography images [P]. United States Patent: US 6819734 B2, Nov 16, 2004.
[22]余晓锷,罗君芳,陈武凡.基于弦图的CT图像环形伪影校正[J].第四军医大学学报, 2009, 30(3):207-209.
[23]罗君方.去除CT图像环形伪影算法研究[D].广州:南方医科大学, 2009.
[24] Sun Shaohua, Zhang Li, Gao Wenhuan. A correction algorithm for detector nonconformity in computed tomography [C]. Nuclear Instruments & Methods in Physics Research A, 2003, 505:552-555.
[25]李保磊,杨民,傅健,等.两种CT成像环状伪影校正方法[J].光学学报, 2009, 29(7):1849-1853.
[26]庄天戈. CT原理和算法[M].第一版.上海:上海交通大学出版社, 1992:1-99.
[27]曹思远.锥束CT三维重建算法加速技术研究[D].重庆:重庆大学, 2009.
[28]谭采云.第三代X线CT环形伪影分析[J]. CT理论与应用研究, 1992, 1(3): 37-40.
[29]朱险峰,何希儒. CT环形伪影原因分析与处理[J].牡丹江医学院学报, 2001, 22(2): 56-57.
[30]庞彦伟,王召巴,林敏. CT图像环形伪影分析[J].华北工学院学报, 2001, 22(1): 16-19.
[31]谭采云.第三代CT环形伪影的分析[J].中国医学装备, 2005, 2(5): 49-50.
[32]刘宾.高分辨率ICT重建技术研究[D].太原:中北大学, 2006.
[33]刘宾,潘晋孝,王明泉.基于错位检测的高空间分辨率ICT重建技术研究[J].科技情报开发与经济, 2006, 26(4): 173-174.
[34]黄战华,蔡怀宇,张以谟,等.基于半像素错位的多幅图像重建高分辨率图像技术研究[J].光学技术, 2002,28(5): 389-394.
[35] Klaus J. Engel, Lothar Spies, Gercon Vogtmeier. Impact of CT detector pixel-to-pixel crosstalk on image quality[J]. Progress in Biomedical Optics and Imaging - Proceedings of SPIE, v 6142 II, 2006.
[36]傅健,路宏年.扇束X射线ICT中环状伪影的一种校正方法[J].光学精密工程, 2002,10(6): 542-546.
[37] Sanjit K.M. Digital signal processing: A computer-based approach[C]. 2nd ed NewYork : MeGraw-Hill.2001.
[38] Rafael C. Gonzalez, Richard E. Woods.数字图像处理[M].北京:电子工业出版社, 2008.
[39] Canny J.A Computational Approach to Edge Detection[J].IEEE Tram PAMI, 1986, 8(6): 679-698.
[40]戴艳.图像边缘检测与应用[D].西安:西安科技大学, 2009.
[41]张震,马驷良,张忠波,等.一种改进的基于Canny算子的图像边缘提取算法[J].吉林大学学报, 2007, 45(2): 244-248.
[42] Jun Li, Sheng Ding. A Research on Improved Canny Edge Detection Algorithm[C]. 2010 3rd International Conference on Computational Intelligence and Industrial Application (PACIIA):68-70.
[43]吴云.光纤面板传像暗影的自动检测[D].大庆:大庆石油学院, 2008.
[44] Gadallah F L, Csillag F, Smith E J M. Destriping multisensor imagery with moment matching[J]. Int. J. Remote Sentsing, 2000, 21(12): 2505-2511.
[45]刘正军,王长耀,王成.成像光谱仪条带噪声去除的改进矩匹配方法[J].遥感学报, 2002, (4): 280-284.
[46]王先培,张爱菊,李少雄.基于非均匀B样条曲线的红外数据的精确拟合及校正[J].光谱学与光谱分析. 2006, 26(10): 2850-2853.
[47]张玲.基于三次样条曲线拟合公路平面线形方法研究[D].武汉:武汉理工大学, 2007.
[48]梁芳.遗传算法的改进及其应用[D].武汉:武汉理工大学, 2008.