复合材料序列图像的分层次配准

Composite serial sections multi-level registration

在复合材料图像三维重构技术中,为了避免直接运用基于特征点的整体配准陷入局部极优,采用分层次的配准方法．首先使用不变矩计算出上下层图像中最相似的颗粒轮廓,然后使用主轴的配准方法完成上下层图像的初步配准,以大幅度减少特征点配准中的优化搜索范围．在计算出轮廓曲线上特征点的基础上,应用最大熵原理和lagrange乘子将点集之间的匹配转化为一个能量函数,再使用最小二乘法计算出使该能量函数值最小的空间变换,得到配准的最优解,从而实现了序列图像的整体精确配准．实验结果表明,本文提出的分层次的配准方法极大地降低了配准过程陷入局部极优的概率,具有较强的鲁棒性和较高的配准精度．

In the three-dimensional reconstruction of composite materials, to avoid local minima during the registration process, a layered registration method was adopted. The most similar grains in two neighboring sections were selected using invariant moments. The preparatory registration was achieved by using principal axis transformation to reduce the searching range. An adaptive method was adopted for the polygonal approximation of the digitized raw contours. The dominant points were identified as the points with local maximum curvatures. A fine point registration algorithm based on maximum entropy theory was adopted. By minimizing the energy function including the joint probability matrix and spatial mapping matrix, the estimation of spatial mapping parameters and joint probability matrix were obtained. The experimental results show the method proposed overcomes the disadvantages that other searching methods easily get into local minima. The algorithm is robust and accurate.