In order to confirm the optimal extraction method for edge spread function (ESF) in knife-edge method, four extraction methods based on the cubic spline interpolation and SG filtering, including Spline、SplineSG、MSG and SASG method are presented. By the experiments using simulation images and remote sensing images, in which only additive noise exists or white noise exists as well at the same time, effects of the four methods are analysed and compared. Experimental results show that Spline and SplineSG method have almost the same results, and the calculating accuracy of line spread function (LSF) and modulation transfer function (MTF) is nearly 2 to 3 times better than SASG method and 3 to 5 times better than MSG method in simulation experiments. While SASG method has the same results compared with Spline and Spline SG method in the practical experiments using remote sensing images. The improving effect of the image quality image restoration is 1.5 to 2 times better than MSG method. When the stripe noise is obvious, SASG method works best. Thus, according to the noise situation, SASG method or Spline method can be chosen for ESF extraction in the implementation process of the knife-edge method.