Structure-oriented edge-preserving smoothing based on accurate estimation of orientation and edges

Abstract

In this paper, we present a new method for reducing seismic noise while preserving structural and stratigraphic discontinuities. Structure-oriented edge-preserving smoothing requires information such as the local orientation and edge of the reflections. The information is usually estimated from seismic data with full frequency bandwidth. When the data has a very low signal to noise ratio (SNR), the noise usually reduces the estimation accuracy. For seismic data with extremely low SNR, the dominant frequency has higher SNR than other frequencies, so it can provide orientation and edge information more reliably than other frequencies. Orientation and edge are usually described in terms of apparent reflection dips and coherence differences, respectively. When frequency changes, both dip and coherence difference change more slowly than the seismogram itself. For this reason, dip and coherence estimated from dominant frequency data can approximately represent those of other frequency data. Ricker wavelet are widely used in seismic modeling. The Marr wavelet has the same shape as Ricker wavelets in both time and frequency domains, so the Marr wavelet transform is selected to divide seismic data into several frequency bands. Reflection apparent dip as well as the edge information can be obtained by scanning the dominant frequency data. This information can be used to selectively smooth the frequency bands (dominant, low, and high frequencies) separately by structure-oriented edge-preserving smoothing technology. The ultimate noise-suppressed seismic data is the combination of the smoothed frequency band data. Application to synthetic and real data shows the method can effectively reduce noise, preserve edges, improve trackable reflection continuity, and maintain useful information in seismic data.

Anisotropic diffusion filtering based on fault confidence measure and stratigraphic coherence coefficients

“…Conventional filtering methods such as Gauss filtering, median filtering and Frequency‐Wavenumber domain filtering etc (Boncelet et al ., 1991; Bakker et al ., 1999; Luo et al ., 2002; AlBinHassan et al ., 2006; Al‐Dossary et al., 2007; Wang Jun et al., 2009; Hale, 2011a; Saleh Al‐Dossary et al., 2016) usually blur the structural information of seismic data while denoising, so it is necessary to find a filtering method that can maintain the edge information of faults and other geologic bodies.…”

Seismic data filtering using non-local means algorithm based on structure tensor

“…Consequently some new techniques, e.g. filtering based on mathematical morphology [6], structure constrained edge-preserved filtering [7], and anisotropic diffusion filtering based on partial differential equation [8], have been developed to preserve edge structures or the contacts between reflections and formations.…”

Noise suppression of time migrated gathers using prestack structure oriented filtering

“…Whitcombe et al (2008) and Helmore (2009) introduce frequency dependent, structureoriented filters. Wang et al (2009) also notice that the local reflector dip depends on the frequency band of the seismic data and suppress the seismic noise using a suite of band-pass structure-oriented filters. Liu et al (2010) reduce random noise, while protecting structural information by combining structural prediction with either mean filtering or lower-upper median (LUM) filtering, using plane-wave destruction filters to estimate the reflector dip.…”