Seismic random noise suppression by structure-oriented BM3D

Abstract

Random noise affects the precision of seismic imaging and interpretation, thus denoising is a significant step in data processing. Block matching and 3D collaborative filtering (BM3D) is an effective block-based noise suppression algorithm. However, when it is applied to the data with complex structures, the similarity between data blocks will decrease significantly, thereby damaging some structural details and leading to energy leakage. To address this issue, we propose a structure-oriented BM3D (SBM3D) denoising method. Initially, plane wave destruction (PWD) is employed to estimate the local slope of the seismic data. Using the obtained slope information, the events are flattened, significantly enhancing the similarity between data blocks. To minimize the flattening error, segmented flattening of the data is performed. Subsequently, BM3D is applied to the flattened data for denoising. Finally, the events are restored to their original shape through inverse flattening to obtain the ultimate denoised result. Through this method, the leaking signal energy from removed noise can be reduced. In addition, we adopt a parallel computing method to improve the computational efficiency. Synthetic and field data testing results show that this improved method can effectively reduce signal damage while ensuring the denoising effect.