Considering the number of the feature points extracted in different images is not equal and their locations are also not corresponding , the scheme presents the closest interval criterion and the svd algorithm to accurately register the retinal images . in order to meet the practical requirements of image registration precision and speed , the paper describes the maximal correlation objective function and uses the simplex - simulated annealing algorithm to optimize the parameter 在基于特征對應(yīng)性圖像配準(zhǔn)算法中,利用多方向探測方法準(zhǔn)確提取了視網(wǎng)膜圖像中的血管交叉點(diǎn)和分枝點(diǎn),對于提取得到的配準(zhǔn)及參考圖像特征點(diǎn)之間非對應(yīng)性且數(shù)目不等這一配準(zhǔn)算法中不可避免的問題,本文提出了距離最近標(biāo)準(zhǔn)和svd分解方法,有效的解決了這一問題。