Seismic Data Denoising Based on DC-PCNN Image Fusion in NSCT Domain

Random noise attenuation is one of the most effective ways to purify effective signals that are used for subsequent seismic imaging and inversion. In order to improve the denoising performance and protect the edge of useful signals, we propose a novel method utilizing an improved pulse-coupled neural network (PCNN) image fusion framework in nonsubsampled contourlet transform (NSCT) domain for seismic signal denoising. First, we use the roughly denoised result based on NSCT and total variation (TV) method as the initial two input data for image fusion. Second, NSCT can transform the two input data into the frequency component in multiscale and multidirection, which can identify more effective signal features. Third, dual-channel PCNN (DC-PCNN) is employed to fuse the coefficients in NSCT domain converted from these two input data, which can achieve a balance between the noise attenuation and useful signal preservation by segmenting and fusing the information. Then, the fused coefficients from different scales and directions can be obtained. Finally, the final fused data can be reconstructed by inverse NSCT to achieve the seismic denoising. The proposed method integrates the advantages of NSCT- and TV-based seismic denoising approaches. Experiments demonstrate that the proposed method obtains the satisfying performance in synthetic and field data examples.