 The proposed method is a course-defined matching strategy that utilizes both global and local information between images to predict corresponding feature points. This method is effective in matching feature points extracted by SIFT or sampled uniformly. It also outperforms other methods in terms of mean-match success ratio and mean-reprojection error. This article was authored by Chen Binliang, Yunyun Dong, Chen Junxiao, and others.