How about filtering in a Fourier domain? If the noise is periodic, then with any luck, your noise will seem to be a pair of good pointed functions in the Fourier space, where you can filter them out with a pair of Gaussians and then convert them back to real space, and your periodic noise should disappear.
source
share