Due to the features of limited frequency band and existing noise for the output images of the conventional 4f information optics system, it is required to protect the image detail regions much better when restoring the images. Based on the classic AWM method, an adaptive window smoothing method which can adaptively adjust the shape of the smoothing window in wavelet domain is proposed to remove the random noise in the system output images. The method utilizes the direction of the high pass subimages decomposed by wavelets to determine the direction of the smoothing window, and adjusts the size of the smoothing window based on the advantage of energy concentration characteristics of subimages in wavelet domain to adjust the shape of the window adaptively. The output image of the 4f system is smoothed with the adjusted windows. The theoretical analysis and experiments show that the method can denoise the image in visual sensation and protect the image detail well, while PSNR is improved about 2 dB. It is suitable for the cases which require a certain image precision and has low pass characters, such as the 4f optic system.