since median works well on Poisson noise, you can use a non local median filter (instead of mean/averaging) for denoising, and use structural similarity etc. to find similar non local patches instead of mean squared error criterion.

well, median filtering was good and much of problem with noise could be solve using this algorithm, but you could experiment with mid point filtering or modified median filtering, there;s a lot of journal that spesific research this, thank you

Alternate algorithms : BayesShrink or SureShrink (using Stein's unbaised risk estimator)

http://link.springer.com/chapter/10.1007/978-3-642-32063-7_34#page-1

http://onlinelibrary.wiley.com/doi/10.1002/ecjc.20357/abstract