For dust impact on machine vision of the probe landing in Mars, a method was brought forward to remove the effect of dust on optical image and provide clear input image for the visual system. First, a model was built for the dust image. Then, the values of the atmospheric light and the transmission coefficient of the model were obtained by calculation. Among them, a quadtree subdivision approach was employed to calculate the value of the atmospheric light. The computing method was to search the maximal average region of the specified threshold area on the minimum image. Then the mean values of each channel on the same position of the original image were calculated as the estimation value of atmosphere light. And on this basis, the calculation of transmission coefficient was performed. At last, the recovery of the clear picture was finished. Test on dust images indicates that the dust image can be restored to clear image by the proposed method. Even in the complex environment, this method has high robustness to illumination variations, dust intensity change and scene change. Compared to other methods, this method has better effect in removing dust impact on optical images and is superior to other methods in terms of the restoring image evaluation index. It can further improve the clearness of dust images and provide more abundant information for post-processing.