We proposed a new iterative technique, which we called the SRP, for restoring SEM images that improves their sharpness. A restored image at each iteration in the SRP is more directly compared with the original image than that in the RL to increase the speed of convergence to the original image. Our simulation results revealed that the SRP sped up deconvolution by about 10–50 times to achieve almost the same improvement in sharpness as that with the RL. We also found that further improvements could be achieved by combining the SRP with the other accelerating method described in Ref. [21]. The experimental results demonstrated significant improvements in image quality.