2-D TFPF based on Contourlet transform for seismic random noise attenuation

• The method is the first to implement 2-D TFPF based on Contourlet transform.
• The method takes the spatial characteristics of the reflection events into account.
• The method improves the performance of TFPF.
• Filtering along a trace matching the event performs better than along each channel.
• The method attenuates random noise a lot and preserves valid signals effectively.

The time–frequency peak filtering (TFPF) algorithm is useful for attenuating seismic random noise. Conventional TFPF processes each channel of the seismic record independently with a fixed window length (WL), which is a one-dimensional algorithm due to filtering along the channel direction. However, the fixed WL is not appropriate for all frequency components at the same time, so using this technique cannot preserve the reflected signals effectively. Also, Conventional TFPF ignores the spatial characteristics of reflection events, resulting in poor continuity of seismic events and serious loss of the correlation among channels. Here we introduce a new spatiotemporal method, called two-dimensional (2-D) TFPF based on Contourlet transform, which considers spatial correlation and improves the performance of the TFPF. Regarding the event as the contour in an image and using Contourlet transform (CT) to the record, we can find the optimal radial filtering trace which best matches the event, and then sample the record to extract signals along the trace. In this way, frequencies of sampled signals are low and similar. After applying the TFPF along the trace instead of along each channel, the estimation bias is decreased due to the low frequency. Moreover, using the same WL is suitable as a result of similar frequencies. Experiments on synthetic models and the field data illustrate that the new method performs well in random noise attenuation and reflection event preservation.