21/22 February 2003Granada iAstro Worshop1 Analysis of Astrophysical Data Cubes using Cross-correlations and Wavelet Denoisings A.Bijaoui 1, D.Mékarnia 1, J.P.Maillard 2, C.Delle Luche 1 1 Observatoire de la Côte d'Azur (Nice) 2 Institut d’Astrophysique de Paris
21/22 February 2003Granada iAstro Worshop2 Outlines The Astrophysical Data Cubes –BEAR and IFTS The Karhunen-Loève expansion (KL/PCA) –The KL basis –The noise of the basis /components Wavelet denoising of the basis/components The residues and their denoising An application on NGC 7027 cube Conclusion
21/22 February 2003Granada iAstro Worshop3 The Integral-Field Spectrographs Different optical devices –Scanning Fabry-Perot –Optical fibers (VIMOS, GIRAFFE) –Cylindrical lenses + Grating (TIGRE, OASIS) –Multislit (SAURON, MUSE) –Imaging Fourier Transform Spectrograph Resulting Data Cubes –Size depending on the device –From Megapixel to Gigapixel Need of specific analysis methods
21/22 February 2003Granada iAstro Worshop4 BEAR : an IFTS device
21/22 February 2003Granada iAstro Worshop5 BEAR at the CFHT focus
21/22 February 2003Granada iAstro Worshop6 The example of NGC 7027 A post AGB planetary nebula –Observations Cox et al –The resampled data cube: 128x128x1024 What information? –Different spectral lines Abundance –Velocity field 3D view –Continuum Necessity to denoise the data cube –To increase the SNR –To observe fainter objects
21/22 February 2003Granada iAstro Worshop7 The data cube
21/22 February 2003Granada iAstro Worshop8 Spectra sample
21/22 February 2003Granada iAstro Worshop9 Elements of the data reduction We can take into account –The cross correlation between the images PCA / KL expansion –The significant details image / image –The significant details spectrum / spectrum Different possible ways –Wavelet Transform + KL exp. + Denoising + Reconstruction (Starck et al. 2001) –KL exp. + Denoising + Reconstruction + Residue + Denoising (Mékarnia et al. 2003)
21/22 February 2003Granada iAstro Worshop10 KL and PCA Search of uncorrelated images The Principal Component Analysis –Iterative extraction of the linear combinations having the greatest variance PCA application to images KL The eigenvalue = the energy / order
21/22 February 2003Granada iAstro Worshop11 The noisy KL basis
21/22 February 2003Granada iAstro Worshop12 Denoising the KL expansion Each KL component is noisy –Depends on the order / eigenvalue Each KL spectrum is noisy The reconstruction from noisy components leads to a noisy restoration Each KL component / spectrum is denoised –Wavelet denoising –Redundant transform –Soft wavelet shrinkage
21/22 February 2003Granada iAstro Worshop13 The denoised KL basis
21/22 February 2003Granada iAstro Worshop14 The residues and their analysis Do not forget to denoise the mean ! The reconstruction with the denoised KL is limited: –Not enough components –Adding components = increase the noise –The denoising can remove local significant feature Use of the residues between the original data and the restored one
21/22 February 2003Granada iAstro Worshop15 After the residue denoising
21/22 February 2003Granada iAstro Worshop16 Spectra Sample
21/22 February 2003Granada iAstro Worshop17 The velocity field
21/22 February 2003Granada iAstro Worshop18 3D visualisation
21/22 February 2003Granada iAstro Worshop19 A spectrum in a cavity
21/22 February 2003Granada iAstro Worshop20 A continuum image
21/22 February 2003Granada iAstro Worshop21 The integrated continuum
21/22 February 2003Granada iAstro Worshop22 CONCLUSION Data cube can be denoised from KL Limitation of the number of components –We could use more components with denoising –Too local information (spectral/spatial) Residue denoising –Could be improved (best basis, softening rule, regularisation,..) Artifact removal –Use of ICA/SOBI blind source separation Help for astrophysical interpretation