三种图像配准法在近红外脉冲光成像中的应用

Applications of Three Registration Algorithms in the Near-infrared Pulsed Light Imaging System

分析了基于特征点控制为基础的配准、傅里叶梅林变换、最大互信息法等三种图像配准算法的原理。采用这三种方法对近红外脉冲光乳腺成像系统采集的病例图像实现图像配准,并从配准所需要时间和最终的配准效果两个方面对三种算法的配准效果进行了比较。结果表明,基于特征点控制基础的配准方法所需时间最短,傅里叶梅林变换次之,最大互信息法所需要的时间最长。但是,从最终配准效果来看,最大互信息法明显优于其他两种算法。应用最大互信息法可以有效实现图像的配准,但是在实际应用中,需要对配准时间进行进一步优化。

The principles of the three image registration algorithms such as the registration based on feature points control,Fourier-Mellin transform and maximal mutual information were analyzed.By implementing these three algorithms to make registrations on all the collected cases,comparisons on the two aspects of registration time and registration effect were made.It is indicated the registration based on feature points control needs the least time,the second is Fourier-Mellin transform,and maximal mutual information needs the most time.But in consideration of the registration effect,the algorithm of maximal mutual information is much better than the other two.Maximal mutual information can achieve the image registration effectively,but in actual applications,especially the registration time needs to be further optimized.