声纳图像配准中的Demons算法

A variational Demons algorithm on sonar image registration

针对传统基于光流场模型的Demons算法变形方向不易确定,同时考虑声纳图像序列中连续帧图像缺乏梯度信息,提出一种结合梯度互信息的改进Demons算法。该方法在原有图像变形力的基础上,增加两幅图像间梯度互信息作为驱动图像变形的附加力。在图像配准的同时,使两幅图像的梯度互信息达到最大,避免Demons算法仅依靠图像灰度梯度变形,从而得到更为精确的配准变换。通过声纳图像配准实验,研究结果表明:该算法在互信息量上提高5%以上,验证该方法具有配准精度高,鲁棒性强,是一种有效的非刚性配准方法。

Considering that a Demons algorithm based on the model of the optical flow field deformation direction can not be determined, and the sonar image sequence lack of gradient information, an improved ＂Demons＂ algorithm was proposed. On the basis of the original image deformation force, the additional force of the mutual information between two images gradient was increased as the driving image deformation. The maximum gradient mutual information of the two images was obtained at the same time of the image registration, preventing the Demons algorithm from relying solely on the image gray gradient deformation, thus resulting in a more accurate registration transformation. Experiments of sonar image are compared with classic Demons algorithm, and the results show that the proposed algorithm increases more than 5% on the mutual information. It has high accuracy, robustness, and is an effective method of non-rigid registration.