Curvelet-domain multiple elimination with sparseness constraints. Felix J Herrmann (EOS-UBC) and Eric J. Verschuur (Delphi, TUD)

Transcription

1 Curvelet-domain multiple elimination with sparseness constraints Felix J Herrmann (EOS-UBC) and Eric J. Verschuur (Delphi, TUD)

2 Context SRME (Verschuur, Guitton, Berkhout, Weglein, Innanen) Sparse Radon (Ulrych, Sacchi, Trad, etc.) Pattern recognition (Spitz) Redundant dictonaries/morphological component separation/pursuits (Mallat, Chen, Donoho, Starck, Elad) 2-D/3-D Curvelets (Starck, Donoho, Candes, Demanet, Ying)

3 Motivation Geometry effects in 3D surface multiple prediction are a bottle neck! New adaptive subtraction robust under phase rotations misalignments incoherent noise Use latest optimal results from theoretical image processing

4 Multiple elimination: denoising /signal separation problem Solve with Curvelets (2-D/3-D) whiten Curvelet spectrum remove multiples with thresholding separate, interpolate & denoise with iterative thresholding Effective when multiple prediction not too far from actual

5 Subtraction with L2 norm Output SRME multi-l2 subtraction

6 Multiple suppression with curvelets Output curvelet filtering with stronger threshold Preserved primaries

7 Surface multiple elimination Multiple prediction: data convolution along surface P 0 (z 0 ) P(z 0 )

8 L2 adaptive subtraction a(t) = Input data with multiples Predicted multiples output data Adaptive subtraction based on minimum energy in the output

9 L2/L1 matched filter denoised {}}{ ˆn Matched filter: : min Φ = d }{{} noisy data p=1 enhances sparseness residue is the denoised data risk of over fitting matched filter {}}{ Φ t m }{{} pred. noise p Guitton& Verschuur 04 May loose primary reflection events...

10 Curvelet domain any 2D data panel decomposition Almost orthogonal decomposition into multiscale basis functions with local frequency and local dip properties Natural basis for wave equations Consist of plane wavelets invariant under convolution

11 Why curvelets Nonseparable Local in 2-D space Local in 2-D Fourier Anisotropic 2 j 2 2j Multiscale Almost orthogonal Tight frame B T B = I Optimal

12 Thanks to Demanet & Ying 3-D Curvelets

13 Thanks to Demanet & Ying 3-D Curvelets

14 Thanks to Demanet & Ying 3-D Curvelets

15 Why curvelets W j = {!, 2 j! 2 j+1, " " J # 2 j/2 } Fourier/SVD/KL f f m F m 1/2, m Wavelet f f W m m 1, m second dyadic partitioning Optimal data adaptive f f A m m 2, m Close to optimal Curvelet source: Candes 01, Stein 90 f f C m C m 2 (log m) 3, m

16 Curvelet Multiples Curvelet Multiple curvelet Almost diagonalize Green s functions (Candes & Demanet 04) Natural basis for wave equations Invariant under convolution, i.e. multiple multiple = curvelet-like Curvelets sense local dip and local frequency content and can discriminate on these properties

17 Curvelet adaptive subtraction Remainder no longer primaries Primaries are model m Multiples are noise n Predicted multiples are ˆn P 0 (z 0 ) m P(z 0 ) ˆn

18 Colored denoising Denoising: noisy data {}}{ d = m }{{} noise-free + col. noise {}}{ n ˆm = arg min m 1 2 C 1/2 n (d m) J(m) with covariance C n E{nn T } and both m, n related to PDE

19 Weighted thresholding Covariance model & noise near diagonal: For ortho basis and app. noise prediction: equivalent to BC n or m B T Γ 2 ˆm = B T S λγ (Bd) ˆm = B T 1 ( d ) arg min m 2 Γ 1 m near diagonal m 1,λ d = Bd, m = Bm and Γ = [diag{diag{bˆn}}] 1/2

20 Filtered input data Multiple suppression with curvelets Bd Curvelet transform Curvelet coeff. data Input data with multiples predicted multiples Threshold Curvelet coeff. pred. multiples Γ Bˆn Inv. curvelet transform Curvelet coeff. primaries S λγ (Bd)

21 Subtraction with L2 norm Input with multiples

22 Subtraction with L2 norm predicted multiples

23 Subtraction with L2 norm Input with multiples

24 Subtraction with L2 norm Output SRME L2 subtraction

25 Subtraction with L2 norm Output SRME multi-l2 subtraction

26 Multiple suppression with curvelets Output curvelet filtering

27 Multiple suppression with curvelets Output curvelet filtering with stronger threshold Preserved primaries

28 Subtraction with L2 norm Input with multiples

29 Time slices L2

30 Time slices curvelet

31 Observations Used 3D SRME to predict noise Can use other noise predictions (e.g. Radon) We do NOT subtract rather mute put to zero or preserve coherent features less sensitive to errors Preserve the edges & primaries!

32 Data from offshore Scotland courtesy Total Stack section with multiples

33 Stack section after L.S. subtraction Data from offshore Scotland courtesy Total

34 Stack section after curvelet-based subtraction Data from offshore Scotland courtesy Total

35 CMP gather with multiples Velocity panel

36 CMP gather after L.S. subtraction Velocity panel

37 CMP gather after curvelet-based subtraction Velocity panel

38 CMP gather after curvelet-based subtraction Velocity panel

39 Iterative thresholding Curvelets are Frames: redundant (factor 7.5-4) thresholding does not solve: ˆm = B T 1 ( d ) arg min m 2 Γ 1 m Alternative formulation by iterative thresholding! m 1,λ

40 Example shallow water environment Input data with multiples True surface multiples

41 Example shallow water environment Input data with multiples True primaries

42 Example shallow water environment Leakage of L.S. subtraction Input data with multiples Result L.S. subtraction

43 Example shallow water environment Input data with multiples Result curvelet-based subtraction

44 Example shallow water environment Input data with multiples True primaries

45 Example shallow water environment Input data with multiples and noise Result curvelet-based subtraction

46 Example shallow water environment! &#!!!! #!!!! &#!!!! #!!! " " # # $ $ % %!"##"$%&'()(&*")+&!,-)".-/#!"#$%&#!'()*$%&*$!"

47 ! 3-D Curvelets! "! #!! #"! "! #!! #"!

48 Conclusions For 3D SRME the acquisition geometry determines the prediction quality and possibilities New domains extend the adaptive subtraction toolbox and hence the quality of the end result Non-linear Curvelet thresholding adds robustness under incoherent noise under missing data Extended to 3-D

49 Acknowledgements Frank Kempe (Cray) for conducting the 3D SRME tests on the Cray-X1 and George Stephenson (Cray) for his support Candes, Donoho, Demanet & Ying for making their Curvelet code available. Partially supported by a NSERC Grant.