Non-local Means (NLM) Filter for Trim Statics Yunsong Huang, Xin Wang, Yunsong Huang, Xin Wang, Gerard T. Schuster KAUST Kirchhoff Migration Kirchhoff+Trim Statics Migration

MotivationsMotivations NLM filterNLM filter  NLM filtering of images  NLM+trim statics of migration images Results (GOM data)Results (GOM data) ConclusionsConclusions Outline

Velocity inaccuraciesVelocity inaccuracies Stacking misaligned prestack migration images  blurred migration images Motivation z x This should be flat. Residual Moveout (RMO) gets a best fit quadratic to flatten CIGs, then we stack to get final stacked migration image dim spots stack zzzz Common Image Gather (CIG)

Motivation CIG 16/31

Motivation CIG 1/31

Motivation CIG 5/31

MotivationsMotivations NLM filterNLM filter  NLM filtering of images  NLM+trim statics of migration images Results (GOM data)Results (GOM data) ConclusionsConclusions Outline

ppbppb Non-Local Mean Filter patch search neighborhood ppappa Filtering (similarity) weight normalizationsensitivity controller Buades et al., 2005 best match survives O a =  W ab I b IbIb OaOa

A noisy image Endo-filtered by NLM filtering weights search neighborhood patch favors repetitive structures favors repetitive structures Non-Local Mean Example

MotivationsMotivations NLM filterNLM filter  NLM filtering of images  NLM+trim statics of migration images Results (GOM data)Results (GOM data) ConclusionsConclusions Outline

B1 B2 A1 NLM filter+ correlate+shift Out of phase Washed out Stacking Strategies Tree representation Poor candidate for pilot stack

1 A 1 2 B 1234 C Recursive Stacking/Destacking Stacking prestack images (no pilot needed) Destacking

MotivationsMotivations NLM filterNLM filter  NLM filtering of images  NLM+trim statics of migration images Results (GOM data)Results (GOM data) ConclusionsConclusions Outline

Stacked Prestack Migration Images Z (km) X (km) 15 0 (31 plane-wave gathers)

Stacked Prestack+Trim Statics Migration Images Z (km) X (km) 15 0 (31 plane-wave gathers)

Z (km) X (km) 15 0 (31 plane-wave gathers) Stacked Prestack Migration Images

Z (km) X (km) 15 0 (31 plane-wave gathers) Stacked Prestack+Trim Statics Migration Images

Too Good to be True Z (km) X (km) 15 0 (31 plane-wave gathers) Really?

Local Trim Statics Shifts * poststackprestack Scatter plot: * x z Cluster mean 33

Ideally, we want a v(x,y,z) that reduces scatter x z Local Trim Statics Shifts

x z over iterations z x 2h 1 2h 2 Relation to subsurface offset Local Trim Statics Shifts RiRi Reduced scatter implies a more accurate velocity model:  = ½  ||R i -R|| 2  MVA

MotivationsMotivations NLM filterNLM filter  NLM filtering of images  NLM+trim statics of migration images Results (GOM data)Results (GOM data) ConclusionsConclusions Outline

Trim statics can align reflectorsTrim statics can align reflectors  mispositioned across prestack images due to velocity inaccuracies Noticeable improvement in feature coherencyNoticeable improvement in feature coherency Limitation: although features are clearly revealed, their locations might still be wrongLimitation: although features are clearly revealed, their locations might still be wrong  We can quantify and reduce the locational variance of the revealed features,  thereby inverting the velocity Conclusions

Thanks to Sponsors of CSIM Consortium

Non-Local Mean Filter patch search neighborhood ppappa ppbppb filtering weight normalizationsenstivity controller Buades et al., 2005 best match survives

Non-Local Mean Filter patch search neighborhood ppappa ppbppb filtering weight normalizationsensitivity controller Buades et al., 2005 best match survives