TEXTURE ANALSYS OF SAR IMAGERY IN THE SPACE-SCALE-POLARIZATION DOMAIN BY WAVELET FRAMES G. De Grandi 1, J. Kropacek 1, A. Gambardella 2, R.M. Lucas 3, M. Migliaccio 2 Joint Research Centre 21027, Ispra (VA), Italy e-mail: frank.de-grandi@jrc.it Università degli Studi di Napoli Parthenope Isola C4-80143 Napoli, Italy e-mail: attilio.gambardella@uniparthenope.it University of Wales at Aberystwyth, Ceredigion, SY23 2AX, UK e-mail: rml@aber.ac.uk
TEXTURE? WHAT TEXTURE? 'Quid est ergo textura? Si nemo ex me quaerat, scio; si quaerenti explicare velim, nescio G. De Grandi, J.S. Lee, D.L. Schuler, "Target Detection and Texture Segmentation in Polarimetric SAR Images Using a Wavelet Frame: Theoretical Aspects", IEEE Transactions on Geoscience and Remote Sensing, vol. 45, no. 11, pp. 3437-3453, Nov. 2007. Augustine as depicted by Sandro Botticelli, c. 1480
CAN WE EXPERIMENTALLY OBSERVE POLARIMETRIC TEXTURE? Eppur si muove (Yet it moves). if I change polarization! Galileo Galilei Portrait of Galileo Galilei by Justus Sustermans (1597-1681), ca. 1639
TEXTURE MEASURES OF POLARIMETRIC SAR DATA BY WAVELET FRAMES [, 2S, S ] HH HV VV WAVELET FRAME MACHINERY S W 2 SPACE (x,y) = f( ψ,s, x, y) s WAVELET ENERGY IN A HYPERSPACE SCALE POLARIZATION STATE
TOOLS FOR TEXTURE MEASURES REPRESENTATION IN A REDUCED SPACE WAVELET SCALING SIGNATURE (WASS) W 2 (x) s = f(s) ( ) [ ] ψ S, 2S, S P = f HH HV Power synthesis at one polarization Detected power data WAVELET FRAME MACHINERY VV highlights the strength of texture and at which scale texture develops. W 2 (y) s = One polarization Two directions (range, azimuth) Dependence on scale f(s)
WAVELET SCALING SIGNATURE (WASS) WASS highlights the strength of texture and at which scale texture develops 2 W log s 2 = f(log 2(s)) 2 A s 1 Non-stationary stationary Indicates stationary nonstationary regime for an underlying scale invariant noise process E γ 2 2H ( ω) ω W s γ = 2H 1 s +
TOOLS FOR TEXTURE MEASURES REPRESENTATION IN A REDUCED SPACE WAVELET POLARIMETRIC SIGNATURE (WASP) WC HV (x, y) Projection of each covariance matrix into the wavelet frame space Scale 8 Covariance matrix representation of each element (x,y) in region of interest C HV ( x, y) WAVELET FRAME MACHINERY Average wavelet energy for each polarization state and each scale WP highlights the strength of texture Scale 4 and at which scale texture develops. WP ( ψ L )(x, y) 1 ψn Polarimetric power synthesis at a set of states 2 s Scale 16 Scale 2 Linear XPOL orientation angle = f(s, ψ Lψ 1 n )
TOOLS FOR TEXTURE MEASURES REPRESENTATION IN A REDUCED SPACE WAVELET FISCHER CLASS SEPARABILITY SIGNATURE (WASEF) Covariance matrix representation of each element (x,y) in region of interest C HV ( x, y) WASP MACHINERY FISCHER SEPARABILITY ( 2 WP ) s Bm = f(s, ψn, x, y,m) R1 ( 2 WP ) B = f(s, ψ, x, y,m) s R2 m Local estimators of wavelet energy of synth power within the two regions Fischer LDA separability criterion highlights the strength of texture and at which scale texture develops. Between regions scatter / Within region scatter LDA = M=8 f( ψ M=4 LINEAR XPOL ORIENTATION ANGLE,s) n n
WASP ANALYSIS OF HIGH RESOLUTION AIRBORNE SAR DATA Theme: Land Cover FOREST BUILDING BARE SOIL E-SAR P-band data set (courtesy DLR)
WASP ANALYSIS OF HIGH RESOLUTION AIRBORNE SAR DATA
WASP ANALYSIS OF SPACE-BORNE C-BAND DATA Marine Applications SIR-C C-band data set English Channel (courtesy JRC EMSL and USGS)
WASP ANALYSIS OF SPACE-BORNE C-BAND DATA
SCALING (WASS) ANALYSIS SAME EXPERIMENT SIR-C C-band data set English Channel (courtesy JRC EMSL and USGS) Polarization orientation: 45 0 Polarization orientation: 45 0 Polarization orientation: 0 0 Polarization orientation: 0 0
SCALING (WASS) ANALYSIS SAME EXPERIMENT Polarization orientation: 45 0 Polarization orientation: 45 0 45 0 Polarization orientation: 0 0 Polarization orientation: 0 0 Polarization orientation: 0 0
WASP ANALYSIS OF L-BAND PALSAR DATA Theme: Swamp Forest Mapping in the Congo Floodplain Primary Rain Forest Degraded Forest Swamp Forst Flooded Swamp Fprest ALOS PALSAR Fine Beam Slant Range Full-Pol Data (JAXA PI Program) Color Composite Pauli Decomposition Image Red: double bounce Green: volume Blue: single bounce
TEXTURAL FISCHER CLASS SEPARABILITY (WASPSEF) Swamp Forest Primary Rain Forest Filtered Wavelet Variance Wavelet Variance ALOS PALSAR Fine Beam Slant Range Full-Pol Data (JAXA PI Program) Color Composite Pauli Decomposition Image Red: double bounce Green: volume Blue: single bounce
BACKSCATTER FISCHER CLASS SEPARABILITY Swamp Forest Primary Rain Forest Backscatter scale 16 Backscatter scale 4 Backscatter scale 1 ALOS PALSAR Fine Beam Slant Range Full-Pol Data (JAXA PI Program) Color Composite Pauli Decomposition Image Red: double bounce Green: volume Blue: single bounce
CONCLUSIONS WEATHER FORECASTS È scherzo od è follia codesta profezia. Is polarimetric texture a prank or madness? Experiments using WASP analysis indicate that indeed texture measures based on wavelet frames Giuseppe Verdi exhibit dependences on polarization state. A Masked Ball However, assessment of the usefulness of these measures in the passage from supervised analysis to segmentation problems in mapping applications still needs to be addressed in a systematic way and in different thematic contexts.