Denoising of Multidimensional Seismic Data in The Physical Domain by a New Non-local Self Similarity Method

Decomposing the seismic data into sparse and low-rank component algorithms is among the applicable methods in efficient seismic random noise attenuation, benefiting from transforming the seismic signal into the other domain rather than the recorded domain. This transformation, however, brings instability and adds the artifact to the data. In the presented study, we propose a novel approach for efficient random noise attenuation by dividing the seismic data (2D and 3D) into several patches based on the characteristic of the seismic signals and the contaminated noise. Subsequently, all the noisy patches are assembled in a predefined analyzing window; the similar patches based on their specific properties are extracted and stacked properly. Then the denoising algorithm is applied to the stacked matrix of the data. Efficient noise separation attains here since the appropriate separation of noise and signal occurs. The full data reconstruction is then applied on the denoised matrix by simply placing desired cleaned columns of the matrix.