基于PCA的无监督群组配准算法研究

Research on Unsupervised Group-wise Registration Algorithm Based on PCA

该研究提出一种基于主成分分析(principal component analysis,PCA)的无监督学习的群组配准框架。由于组平均算法存在迭代时间较长且创建的模板较为平滑的问题,进而提出PCA算法构建模板。利用PCA改进的权值计算的方法重新构建参考图像,将构建的参考图像和输入图像送到无监督网络进行配准。实验结果证明,提出的算法在配准精度有待改善,但在模板构建的收敛速度以及所构建的模板平滑性上表现更好。

This study proposes a group registration framework for unsupervised learning based on principal component analysis(PCA).Due to the problem that the group average algorithm has a long iteration time and the template created is relatively smooth,the PCA algorithm construction template is proposed.the reference image is reconstructed by using the improved weight calculation method of PCA,and the constructed reference image and input image are sent to the unsupervised network for registration.Experimental results show that the proposed algorithm need improve the accuracy of registration,but performs better in the convergence speed of template construction and the smoothness of the constructed template.