In Fig. 10, we demonstrate the performance of the convergence of variance technique for the three target types in the presence of Poisson noise. Using the RL deconvolution algorithm with a cap on the number of iterations set to 10,000 and an MCFA consisting of 30 frames, we were able to recover the correct in all cases. For the point source targets, the blurring effects of the simulated atmosphere reduce the intensity to the point that it is difficult to visually identify the various point sources. However, when deconvolution is completed, each of the sources is easily identified. For the cross bar pattern, the process of deconvolution makes it much easier to identify the structure of the target pattern. Although it may seem that a cap of 10,000 iterations is unreasonable, the algorithm will only take as much time as needed to converge within this upper bound. For instance, if the termination criteria is met before the upper bound on iterations is achieved, the algorithm will terminate. Even with a standard home computer with a 2.7 GHz Intel® Core™ i5 processor and 16 GB of memory, convergence was achieved in a reasonably short period of time for all target types as shown in Table 7. Again, as indicated in Fig. 5, the algorithm could be manually interrupted sooner if needed, at which point, the lowest value of that has allowed convergence would be returned as the answer. For instance, in the example shown in Fig. 10, if we interrupted the routine after 4.26 s for the cross bar pattern, we would get an estimated of 4 cm. After a total of 10 s, the estimate would be within 0.5 cm of the truth at 3.8 cm. The estimate continues to be refined until the best estimate is achieved after 21.77 s. As demonstrated in Table 6, convergence at a value less than the truth is not an issue provided we have adequate SNR, and the image is not further degraded by optical aberrations such as focus error.