基于多分辨率和Powell算法的医学图像配准

Medical image registration based on multi-resolution and Powell algorithm

针对基于互信息和Powell算法存在局部收敛的问题,在改进的Powell算法的基础上,提出了一种基于多分辨率策略的医学图像配准算法。首先通过小波变换对源图像进行分层,然后在最低频带使用改进的Powell算法进行搜索,并利用搜索结果来指导上一层的搜索,逐层细化,由粗到细,最终实现图像的精确配准。实验结果表明,该方法较传统方法速度快、精度高、鲁棒性好,同时能有效避免局部收敛。

On the basis of the modified Powell algorithm,this paper presented an image registration algorithm depending on multi-resolution strategy,which aimed to solve the problem of local convergence based on mutual information and the Powell algorithm.First,it decomposed wavelet sub-bands of the original images using the wavelet transform.Then it used the modified Powell algorithm to search in the lower band,and guided the previous layer search by the search results.It refined layer by layer,from coarse to fine,achieved the image registration ultimately.The experimental results show that the method is faster,higher precision and better robustness than the traditional methods.And further more,it can avoid the local convergence effectively.