Posterior Sampling for Random Noise Attenuation

What is it about?

For the problem of random noise attenuation, we give a method to sample stochastic solutions from the posterior distribution, which is obviously different from other methods that give a deterministic solution. Every stochastic solution is reasonable and of high quality. We also achieve interactive posterior sampling by automatically estimating the noise level or manually setting it according to a noise level map of the field noise, and accordingly we give three modes of noise suppression: mild, moderate, and strong. Experiments on synthetic and field data show that the posterior sampling using SGMs can more effectively recover the details of the useful signal while attenuating random noise compared to different non-DL/DL methods. In addition, experiments on interactive posterior sampling show that interaction with users provides flexible parameter selection similar to traditional methods and provides more diverse solutions.

Featured Image

Photo by julien Tromeur on Unsplash

Photo by julien Tromeur on Unsplash

Why is it important?

We give a method to sample stochastic solutions from the posterior distribution, which is obviously different from other methods that give a deterministic solution. Every stochastic solution is reasonable and of high quality.

Perspectives