Advancements on the use of the non local means algorithm for seismic data processing

Among the techniques and algorithms usually applied to seismic data for random noise attenuation, Non Local Means (NLM) filtering is a promising option. It is based on a weighted mean in which the weights depend on the measure of similarity between patches surrounding each sample. This methodology allows the preservation of features and structures while incoherent signal is filtered out. The application of such methodology can be n-dimensional but its high computational cost disadvantages a 3D implementation. In previous work we presented a revised version of the NLM algorithm, improved from both the computational and signal to noise enhancement points of view. In the present paper we focus on the application of this revised NLM on real data time-slices, and investigate more in detail the 3D implementation of the method.