Phase diversity (PD) is a kind of wavefront sensing technique based on the measurement of two or more object images. The optical system involved is relatively simple. Using the methods of optimization and image processing, it can jointly estimate the phase aberration as well as the object itself simultaneously. However, when additive noise appears in the images, many local minimum are produced in the function function. Consequently, the convergence of the algorithm is influenced, or even worse, the algorithm will be trapped in the local minimum which are far from the global minimum. It is necessary to preprocess the image to lower the affection of noise before using the method of optimization. Low pass filter is chosen to eliminate the noise because the images contain less high-frequency components than the lower-frequency ones due to the finite size of the pupil of the optical system. The Butterworth low-pass filter is chosen after considering the performance of some low-pass filters. The research results demonstrate that the speed of convergence and precision of the algorithm is dramatically improved when the filtered images are used, and the RMSE between the estimated and real object is less than before.