摘要
提出了一种新的医学切片图像配准方法,该方法分两步完成:首先引入图像归一化相关理论对传统的主轴矩配准算法进行改进,并用改进后的主轴矩法对图像进行粗配,得到初始变换参数,该参数可作为精配准的初始值;然后在初始值附近,运用修正后的powell算法循环迭代使得两幅图像的互信息最大,此时得到的变换参数可实现更高精度的配准:最后在MATLAB上仿真,实验结果表明,利用主轴矩法对图像进行粗配,减少了后续配准的搜索步骤,提高配准效率,互信息法提高了配准的精度。
A new image registration method was proposed in this paper, It completed in two steps. Firstly, Quote image normalization theory to the traditional principal axis algorithm for improving, use the improved principal axis algorithm to complete rough registration and get the initial parameters. Then, Iteration of the loop by the modified Powell algorithm in the vicinity of the initial value to maximum the the mutual information of two images, and use the new transformation parameters to complete a more accurate image registration.Finally, Simulation in the MATLAB. experiment, The results showed that the firstly reduced the number of the followed steps, and the Mutua~ information method improved the accuracy of the registration.
基金
广州市科技计划项目12C22111580
广东省自然科学基金项目9151008901000006
关键词
互信息
图像归一化
神经切片图像
Image registration Mutual information Image normalization Nerveslice image
