In order to evaluate the point spread function (PSF) of optical remote sensors well and truly, a fast multichannel blind deconvolution (MBD)-based estimation method which does not need any ground target is proposed. Multiple sub-images with uniform local background are extracted from a remote sensor image and alternate minimization algorithm is used to implement the blind deconvolution. It only takes 0.4917 s to achieve a percentage mean square error of 1.1% with two noise-free sub-images on a Matlab platform. While the error is 5.9% for two sub-images with the signal-to-noise ratio of 45 dB. By applying the estimated PSF to image restoration, it can bring gray mean gradient from 5.7 to 7.1 and energy of Laplacian from 29 to 46. This method performs more accurately and efficiently than the frequently used slanted-edge method.