A new method of getting transformation among two images by feature point matching was proposed to match the serial dynamic images fast and exactly. First,in the light of multiple restriction (MR) criterion,including local information entropy,similarity measure and criterion of constant distance scale,three pairs of matched points were found exactly between two images. Then,using the points,six affine parameters of two images were computed by matrix LSM. Experimental results indicate that compared with correlation method,MR algorithm can wipe the false matching points off resulting from the movement of object. Aiming at 384 pixel×256 pixel images,only 2.76 ms is needed for MR algorithm to complete feature points matching. 58% time was reduced. And the error of affine parameters estimating in X,Y direction,was cut down to Δx=0.13,Δy=0.02,greatly less than one pixel. This method can meet the needs of high speed,precision and adaptability in engineering field.