论文研究 车辆导航中的交互多模卡尔曼滤波跟踪方法.pdf

图像的刚性配准或者匹配是图像处理中的热点和难点问题。由于图像的不连续性和成像设置中噪声的影响，传统的最小二乘方法（Least Squares Method）不能很好地解决这一难题。采用M估计算子（M-estimation operator）代替最小二乘算子，并结合由粗到精的多分辨率策略，提出了一种新的基于M估计的医学图像多分辨率配准算法，实验结果表明该方法对配准的速度、精度及鲁棒性都有很大的提高。2062009,45(14)Computer Engineering and Applications计算机工程与应用其中CT图像100幅、MR图像100幅。图3为实验中所用的部5结语分图像,第一行为MR图像,第二行为CT图像。分别对它们进采用M估计算子代替最小二乘算子,并结合由粗到精的行相同的处理,先进行旋转,再沿x,y轴分别进行平移。旋转角多分辨率方法对不同模态的医学图像进行了实验,从实验结果度和平移量是随机产生的。如果计算岀的变换参数与真实值相可以看岀,本文算法能够准确、快速地处理刚性配准问题,特别差1个像素或1度,即认为配准成功。本实验中旋转角度和平适用于医学图像的配准。但该方法对数据的缺失及图像变形较移量满足[-20,20均匀分布。利用本文算法和最小二乘法分别敏感,对于这一类情况还需进一步研究讨论,以改善该方法的进行配准实验,结果如表I所示。通用性。参考文献:l1] Maintz J B A, Viergever M A. A survey of medical image registraion[J]. Med Image Anal, 1998, 2(1): 1-362 Pluim J, Maintz J, Viergever M Mutual information based registration of medical images: A survey[.IEEE Trans Med Imag, 2003, 22(8):986-10043 Brown L G.a survey of image registration techniques [J).ACMComput Surv,1992,24(4):325-376[4] Gonzales R C, Woods R E Digital image processing M] .2nd ed Up-per Saddle River, NJ: Prentice-Hall, 2002图3实验中所用部分图像[5 Black M J, Anandan. A framework for robust estimation of opticalflow[C]/Proceedings of the International Conference on Computer表1配准结果Ⅴ IsIon(ICCV),1993:231-236总和时间minNestares 0, Heeger D J Robust multiresolution alignment of MRI实验数量100100200brain volumes J). Magnetic Resonance in Medicine, 2000, 43: 705-715本文算法99%92%95.5%[7 Li S Z Markov random field modeling in image analysis [M].2nd最小二乘81%78%79.5%ed. [S.L. ] Springer-Verlag, 2001[8Li S Z, Wang Il, Soh w Y C Robust estimation of rotation angles从200组对比实验的实验结果可以看出,本文算法实现from image sequences using the annealing M estimator [J]-Journal了总体95.5%的准确率,而最小二乘法的配准效果较差,准确of Mathematical Imaging and Vision, 1998, 8(2): 181-192率只有为79.5%,并且本文算法所用时间仅为最小二乘方法9ButP. Adelson h. The laplacian pyramid as a compact image的1/3code[J.IEEE Trans Commun, 1983, 31: 532-540(上接190页)3 Kybic J, Unser M Fast parametric elastic image registration[J.IEEE量函数之间的关系,虚线表示最速下降法的迭代次数与能量函Trans Image Process, 2003, 12(11): 120-128.数之间的关系。从图4可以看出 Levenberg- Marquard优化方法4 Unser M, Aldroubi a. fast b- spline transforms for continuous im-加大了收敛速度,节省了配准时间,从而使医学图像配准能够age representation and interpolation[J.IEEE Transactions on Signal在很大程度上满足临床实时需求。Prucessilly,1991:821-8325 Unser M, Aldroubi A, Eden M B-spline signal processing: Part IIJIEEE993:821-832.4总结本文应用 Levenberg- Marquard方法对图像进行配准的同16) Zitova b, Flusser J Image registration methods: A survey [J]. Imageand Vision Computing, 2003, 21: 977-1000时采用基于B样条的变形模型以及SD标准作为代价函数,[7 Unser M, AldroubiA, Eden M the2 polynomial spline pyramid[J传统基于B样条的变形模型计算速度较慢,很难适应临床应用IEeE Trans Pattern Anal Mach Intell. 1993. 15(4):364-379本文的卷积计算提高了计算速度;应用最速下降法虽然易于操[8] Unser M, Aldroubi a, Eden mb- -spline signal processing:Part2作,但要经过多次迭代才能收敛,对此问题本文应用了 LevenEfficient design and applications[J]. IEEE Transactions on Signalberg- Marquard方法,以提高配准速度和精度;对图像配准进行Processing, 1993 834-848优化时较强烈的变形很难恢复到理想的效果,在配准过程中充 ohlfing T, Maurer C R, Bluemke D A,eta. Volume preserving分应用了先验知识对代价函数进行了修正,以得到配准参量的non-rigid registration of MR breast images using free-form defor全局最优解,得到平滑、连续全局最优的变换,最后得到精确的mation with an incompressibility constraint(J.IEEE Trans Med, Im-配准结果。实验结果验证了该方法的有效性。age,2003,6(22):730-741.[10] Stefanescu R, Pennec XGrid powered nonlinear image registrationly adaptive registration[J]. Medical Image Analysis, 2004参考文献:(5):243-278[I] Lim J, Yang M Direct method for modeling non-rigid motion with [11] Pilet J, Lepetit V, Fua P Fast non-rigid surface detection, registrathin plate spline[c]/Proe CVPR, 2005tion and realistic augmentation[J). International Journal of Computer[2 Rueckert D, Sonoda L l, Hayes C, et al. Non-rigid registration usingVision 2007free-form deformations: Application to breast MR images [J].IEEE [12] Szeliski R A comparative study of energy minimization methodTransaction on Medical Imaging, 1999, 18(8): 712-721for Markov random fields[C/Proc ECCV, 2006