iitl-h. '^r^^-' LINEAR FEATURE DETECTION IN SAR IMAGES Ph. COURMONTAGNE L2MP UMR CNRS 6137 / ISEM L2MP UMR 6137
Report Documentation Page Form Approved OMB No. 0704-0188 Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it does not display a currently valid OMB control number. 1. REPORT DATE 14 MAR 2001 2. REPORT TYPE N/A 3. DATES COVERED - 4. TITLE AND SUBTITLE Linear Feature Detection In SAR Images 5a. CONTRACT NUMBER 5b. GRANT NUMBER 5c. PROGRAM ELEMENT NUMBER 6. AUTHOR(S) Ph. COURMONTAGNE 5d. PROJECT NUMBER 5e. TASK NUMBER 5f. WORK UNIT NUMBER 7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) ISEM Laboratoire L2MP UMR 6137 CNRS Maison des Technologies Place Georges Pompidou 83000 TOULON FRANCE 8. PERFORMING ORGANIZATION REPORT NUMBER 9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR S ACRONYM(S) 12. DISTRIBUTION/AVAILABILITY STATEMENT Approved for public release, distribution unlimited 13. SUPPLEMENTARY NOTES See ADM001263 for entire Adaptive Sensor Array Processing Workshop. 14. ABSTRACT See Briefing Charts 15. SUBJECT TERMS 11. SPONSOR/MONITOR S REPORT NUMBER(S) 16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF ABSTRACT UU a. REPORT unclassified b. ABSTRACT unclassified c. THIS PAGE unclassified 18. NUMBER OF PAGES 18 19a. NAME OF RESPONSIBLE PERSON Standard Form 298 (Rev. 8-98) Prescribed by ANSI Std Z39-18
INTRODUCTION Ship wakes in Synthetic Aperture Radar (SAR) images Detection using Radon transform (DEANS, 1983) Straight line Radon transform Peak or trough SAR images affected by speckle Detection wedding between the Radon transform and a filtering method 9th ASAP Workshop 13/14 March 2001 1/17
CONTENTS Radon transform discrete form Stochastic matched filtering method Interpolation-filtering method theory and subimage processing Experimental results on SAR images comparison with the classical approach 9th ASAP Workshop 13/14 March 2001 2/17
3 LINEAR FEATURE DETECTION IN SAR IMAGES THE RADON TRANSFORM Consider an image I, with dimensions The Radon transform where MI2 of this image is: (xe, y^) Z and ^ e [0; n] Problems : - new pixel coordinates in Computation of the new pixel values - pixels localized at the edge in (surrounding points) Edge effect size of the image in : 9th ASAP Workshop 13/14 March 2001 3/17 M xm I{xs: B) ^ I{x& COS 0 He sin 9, XQ sin 9 -hys cos 0) yg = -M/2 R^t 5Rfl^/.0*^,- ''' Ue ^ye h -.. 0 < ri 9 n ;7T^? >,- ^' v li * V ". ^ _. V ^jf _,.r' \ - -'. V < ':, _^ C. '.' \ -^ _, ' y' _\ 5" V" en M M V2 ".V.' ' \ ^. ',' 1^ _,._ ^^. vt j-o V - ' ^^ '(^ ^^r., - 1 V * ' i-',\ _.i \ ^h h ^,»- \ '.-_.-= : '- \ ' ' ^ ' \ u \ V2.7;
STOCHASTIC MATCHED FILTERING METHOD Consider the stationary noise-corrupted signal defined over D : where signal S{x,y) and noise B{x,y) are assumed to be independent. Observed signal expansion: Z{x,y) = S{x,y) + B{x,y) Z{x,y) -^2 *n(a;,?/) n=l : deterministic and linearly independent basis functions where: Q : number of basis functions retained for the expansion uncorrelated random variables defined by Zfi // Z{x,y)^n{x,y)dxdy 9th ASAP Workshop 13/14 March 2001 4/17
Determination of functions ^n{x, y) Making sure n are uncorrelated Optimization of the signal to noise ratio K expressed as a Rayleigh quotient K will be maximized if ^n{x, y) are solutions of: jj Tss{^-x',y-y')^n{x\y')dx'dy =\n 11 TBB{X-x,y-y')^n{x',y')dx'dy' where ^ss and ^BB are the covariances of the signal and of the noise Determination of functions *n{x,y) * {x,y)= // YBB[x-x',y-y')^n{x\y')dx'dy' Signal to noise ratio of the component of : number Q of basis functions such as Ag>l 9th ASAP Workshop 13/14 March 2001 5/17 n th Z(x,y) -4X ^s 2 ''n ^B
INTERPOLATION-FILTERING METHOD Basic idea Observed signal signal Z{x, y) ^n{x,y) Ki^.y) Zn and andq Mx,y) ^^Zn'^iix-.y) n=l Observed signal expansion and restoration using interpolated functions *»(^,y) Defaults: heavy CPU budget and memory problems new formulation using the discrete cosine transform 9th ASAP Workshop 13/14 March 2001 6/17
Z{x, y) interpolation-filtering using DCT, with D = [-T; T] Functions DCT coefficients *n(a:, y) <l *n(x,?/) Z{x, y) Z{x,y) ^ss ^BB an yp,q ^p,q dk,i ^\^ll,q ^%l,p,q ^n{x, y) <J=^ *n(x,y)<^=^ 'n 1 T Z{x, y) '\ r Nf Nf Nf A'/ Af/ Nf Nf p^o q=0 Q Nf Signal of interest restoration in 3fJ e Nf Nf Zeixe^ye) = >. >.t'fc^^cos I r^i^ 1 cos 2T fc=0 /=o 9th ASAP Workshop 13/14 March 2001 7/17 2T
Subimage processing Assumption: Z{x,ij) is stationary necessity of a subimage processing Edge effect overlapping between adjacent subimages: (/- A^ cos ^ + sin 6* and h = N sin e sin 9 + cos 0 Smoothing effect when Q is the same for the whole image minimization of the mean square error to find 9th ASAP Workshop 13/14 March 2001 8/17 Q for each subimage Q Nf Nf 1 ^-=-?- + iek-ax)ee(/5m)
EXPERIMENTAL RESULTS SIR-C/X-SAR image moving ship dark turbulent wake (17 miles) 256 gray levels (0: black, 255: white) Image size: 698 X 698 Variation coefficient: 0.277 300 200 Z>0 Difficulty: dark patches in the upper right corresponding to smooth area of low wind 9th ASAP Workshop 13/14 March 2001 9/17
Signal and noise auto-correlation functions a priori knowledge of the signal and the noise auto-correlation functions determination of normalized auto-correlation models Signal auto-correlation function 9th ASAP Workshop 13/14 March 2001 10/17 Noise auto-correlation function
Interpolation-filtering of the SIR-C/X-SAR image 1 r 1 1 1 1 r 1 1 50 Rotation angle: CO 35 400 XO ^ - K)0 Number Q of basis functions: approximately 13 for the wake 200 100 ~ 0 > - 150 near 1 for the rest of the image loo > - - loo 200 V - Image size: 973 X 973 300 ' 1-400 Variation coefficient: 0.016 1 1 _i 1 j 1-400 -3C0 -ECC -lo) 0 ico 200 300 400 SIR-C/X-SAR image in reference system 5R 35 9th ASAP Workshop 13/14 March 2001 11/17
Radon domain displayed as image <^ Angles -30CO Angles -90CO -fioco -SKO -400-300 '3JD 'lod D 'DD 300 300 «0 Projections Our processing Trough corresponding to the wake localized near 9th ASAP Workshop 13/14 March 2001 12/17 5r Projections Classical processing
Radon domain displayed as graph 6000 xlo^ 4000-2000 Amplitude '2000-4000 Amplitude -MOO -aooo -10000 Angles Our processing Angles Classical processing No ambiguity about the presence of a ship wake with our processing detection by simply using a threshold contrarily to classical approach 9th ASAP Workshop 13/14 March 2001 13/17
Reconstructed image using the transform domain Inverse Radon transform applied to the transform domain raised to the power of 3 -S^l fu -3ni?» SfO 3in 21X1 I5D -im 15(1 lua ICO IK «30] -im -jro -ifyi CPH mi iro Our processing Variation coefficient: 0.002 Classical processing Variation coefficient: 0.145 9th ASAP Workshop 13/14 March 2001 14/17
ANOTHER SAR IMAGE ERS SAR image ship pixels replaced by mean background dark turbulent wake 256 gray levels (0: black, 255: white) Image size: 325 X 325 Difficulty: dark patch near the wake (slick oil) 9th ASAP Workshop 13/14 March 2001 15/17
]000 LINEAR FEATURE DETECTION IN SAR IMAGES Radon domain displayed as image f ma Qi Angles n m I Qi Angles * 1 10O3 M ^M-^m ^M-m> ^M^}m ^M-m> 140 H-soco m * ' ^ ^' ' '' ' -20} -t50 -TO) -50 0 ffi 'DD 'K??00 Projections Our processing?» '150 -K -SD 0 a ^w»?:c Projections Classical processing ^ -60DD -TODD Trough corresponding to the wake localized near 115^ 9th ASAP Workshop 13/14 March 2001 16/17
CONCLUSION New processing for ship wakes detection SAR image Radon transform original contribution: taking into account the speckle for image interpolation Better detection with lower probability of false alarm or no detection Future work: extension of this method to the localized Radon transform 9th ASAP Workshop 13/14 March 2001 17/17