Phase diversity technique (PD) is a widely known method to estimate wave-front aberration of optical imaging system and to obtain reconstructed high-resolution image after degradation. However, when detecting weak or low light object in space, Poisson noise, as the main source of noise, has a serious impact on the accuracy of the PD's two main function. Hence, we firstly propose a modified PD combined with Non-local Means (NLM) algorithm to reduce the sensitivity of PD towards the Poisson noise. The numerical simulations demonstrate that our approach compared with the traditional PD has a significant improvement in terms of both the wave-front residual root-mean-square error (RMS) and the structural similarity index metrics (SSIM). The wave-front residual RMS decreases by approximately 51% across the Poisson noise levels ranging from 24.48 dB to 61.02 dB. The overall decline range of SSIM significantly decreases from 47% to 17%, and the average of SSIM increases from 84% to 91%. The modified PD would be useful in the AO system with active Poisson noise. © 2019 COPYRIGHT SPIE. Downloading of the abstract is permitted for personal use only.