在磁共振脑功能成像 (functionalmagneticresonanceimaging ,fMRI)试验中 ,头部很小的移动就会对试验结果产生不良影响 ,因此在对数据统计分析之前要求进行移动校正 .本文利用LMF(levenberg mrquardt fletcher)算法极小化参考图像和校正图像之间的残差平方和自动校正fMRI图像 ,根据多分辨率金字塔结构由粗到细地搜索最优解 ,既可以较快的收敛 ,又可以考虑较大的移动 . In the functional magnetic resonance imaging (fMRI) test, a small head movement has an adverse effect on the test result, so motion correction is required before the statistical analysis of the data.This paper uses the LMF (levenberg mr quardt fletcher) algorithm FMRI images are automatically corrected according to the sum of squared residuals between the minimized reference image and the corrected image. According to the multi-resolution pyramid structure, the optimal solution is searched from coarse to fine, which can not only converge faster but also consider the larger movement .