Abstract: A new elastic image registration algorithm is presented in this article.The article utilizes B-spline deformation model，because the B-spline has a second-order differential and nuclear convolution can be calculated in a separate identity，the convolution calculation based on the B-spline convolution of the nuclear，so the speed of calculation is fast.At the same time，because of the full application of optimization methods on the control factor of Hessian matrix，the deformation is more smooth；the cost function applies SSD.The method not only improves the speed but also strengthens the robustness of the registration.To be smooth and stable the offset，the application of the prior knowledge of the cost function is amended so as to offset bound to carry out.

Key words: elastic registration, B-splines, Levenberg-Marquardt optimal method, convolution

摘要： 提出了一种快速医学图像弹性配准计算方法。该方法应用B样条变形模型，由于三次B样条函数具有连续二阶微分的特性并且可以单独计算，该方法采用了基于B样条卷积核的卷积计算方法，提高了计算速度。同时该方法应用Levenberg-Marquardt优化方法计算控制系数，此优化方法充分应用关于控制系数的Hessian矩阵，使形变更为平滑。使用卷积计算及Levenberg-Marquardt优化方法不仅有很高的计算速度，也有很强的配准鲁棒性。为了得到平滑稳定的偏移量，该方法应用先验知识对代价函数进行了修正，从而对偏移量进行约束。

关键词: 弹性配准, B样条, Levenberg-Maquardt方法, 卷积