Classification of high resolution sar images using textural features

8/6/2019 Classification of high resolution sar images using textural features

1/24

Aurlie Voisin, Vladimir A. Krylov, and Josiane Zerubia


2/24

Synthetic Aperture Radar (SAR): active sensor used fordifferent kinds of applications: Security, Epidemiology, Environment, Risk management

Very high resolution (VHR) SAR data (up to 1 meter) Speckle [Oliver 04]

Heterogeneity [Cheney 09]

Supervised classification of VHR single-channel SARamplitude images Methods : neural networks [Jacob 02], bags-of-features [Yang 09], etc.

Proposed method: statistical modeling + Markov random fields (MRFs)

2


3/24

3

Supervised classification

1 learning image

1 test image

M classes (e.g.: water, vegetation, urban)

Original SAR

amplitude image


4/24

4

Learning

Test

PDF SAR amplitude of

each class

MRF parameter

estimation

Energy minimization to

find the class of the test-image pixels


5/24

5

Learning

Test


each class

MRF parameter

estimation




6/24

Dictionary-based Stochastic Expectation Maximization

Heterogeneity of SAR images

Single probability density function (PDF) doesnt accurately model

SAR amplitude statistics

Model the SAR amplitude PDF by a FMM (Finite Mixture Model)

6

K

k

kkk rpPrp1

)(.)(

[Moser 10]


7/24

Estimation of:

K: number of components

pk(r| k): kth PDF family component

Parameters of the kth component (k and Pk)

with

Data incompleteness

belongs to which statistical population?

Unsupervised context

7

K

k

kP

1

1 10 kP

Nrrr ,...,1

K

k

kkk rpPrp1

)(.)(


8/248

Log Normal

Weibull

Fisher

Generalized Gamma

Nakagami

K-Root

Generalised Gaussian Rayleigh

Heuristic

Theoretical

2

))(ln(

12

1

),(

mr

ermrf

r

errf1

2 ),(

ML

L

MLr

M

Lr

M

L

ML

MLMLrf

)1(

)(

)()(

)(),,(

1

3

r

er

rf

1

4)(.

),,(

12

5)(

2),(

Lr

L

L

erL

LLrf

2/1

12

6 2)()(

4),,(

LMrKr

LM

MLMLrf LM

ML

ML

der

rfr

2/

0

)sin()cos(

7

/1/1/1

)(

),(

09/21/2010 SPIE Remote Sensing Toulouse 2010 - A. Voisin et al.

[Moser 06]

[Krylov 09]


9/24

E step:

S step: stochastic labeling step leading to

M step:

9

K

l

t

lil

t

l

t

kik

t

kt

ik

rpP

rpP

1

),(.

),(.

t

ikw

N

i

kik

t

ik

t

krpw

k 1

1),(lnmaxarg

imagesize

kpixelsnbP

t

k_

_1

[Celeux 95]


10/24

ML not feasible for some PDF distributions

Method ofLog-Cumulants [Moser 06, Tison 04]:

Mellin transform of the PDF [Sneddon 72]

th order second kind cumulant

10

01).())(()( duuupspMs suuu

)1())(ln()(v

uv Family MoLC equations

Log-Normal 1=m, 2=

Weibull 1=ln()+(1)/, 2=(1,1)/

Nakagami 21=(L)-ln(L/), 42=(1,L)

Generalized Gamma 1=(k)/+ln(), 2=(1, k)/

3=(2, k)/3


11/24

E - step: (posterior proba)

S step: sample the label for each greylevel according to

MoLC step:

K step: for k=1,,Kt, if Pk

t+1 < threshold, eliminate the kth

component. Kt+1 = Kt-1

Model Selection step: for each k, compute the log-likelihood,

and define pkt+1(.) as the PDF yielding the highest value

11

tK

l

t

l

t

l

t

l

t

k

t

k

t

kt

k

zpP

zpP

1

),(.

),(.

)(zst

ktQz

t

kjj

t

kj zfzhL )(ln).(

t

k

1,...,0 Zz

1

0

1

)(

)(

Z

z

Qzt

k

zh

zh

P kt

kt

kt

Qz

Qzt

kzh

zzh

)(

)ln().(1

1

kt

kt

Qz

Qz

bt

k

t

bkzh

zzh

)(

))).(ln(( 1

kzszQt

kt )(:

[Moser 06]

[Krylov 09]

3,2b


12/24

Initialization Kmax = 6 chosen by trial and error

Randomly chosen labels

For each iteration t, the global log-likelihood is

computed, if it is > (max log-likelihood), the parameters

are saved

Stopping criterion

Maximum number of iterations reached

Number of components = 1

12


13/2413

Model KS distance

GGamma 0.022

DSEM 4: GGamma,

Naka, LogN, Weib.

0.007

TerraSAR-XimageofRoshe

nheim

(Germany)(Infoterra,2008)

COSMO-Skymedimageof

Cavallermaggiore(It

aly)

(ISA,2008)

Model KS distance

Weibull 0.053

DSEM 4: LogN,

Naka (2), Weib.

0.011

Statistics of the

whole imageStatistics of the

vegetation class


14/2414

Learning

Test


each class

MRF parameter

estimation




15/2415

MRF: robustness against speckle and contextualinformation [Besag 86, Dubes 89, Fjortoft 03]

Anisotropic second-order neighborhood system

Gibbs distribution[Besag 74, Geman 84]

: Localcharacteristic (conditional proba) for each class m

Potts model:

Pseudo-logLikelihood

M;1

M

j

xxH

xxH

s

s

mss

ms

s

mss

js

sms

e

e

xP

xxPxxPxp

1

),(

),(

)(

)(

)()(

)(

)(

,)()(

Ss Csss

xxs

s ssPDFxxH

',:'

)(

'.)ln(),,(

scliques

s

ms xpxPL_

)()(ln))(ln(


16/2416

Problematic: Single pol. CSK images

Find a 2nd channel to improve the accuracy

Textural features

COSMO-Skymedimageof

Cavallermaggiore(Ita

ly)(ISA,2008)

Semi-Variogram [Chen 04]

Grey-Level Co-occurrence

Matrix (GLCM) variance

[Haralick 73]


17/2417

PDF SAR amplitude ofeach class and each

channel

Copulas

MRF parameter

estimation


find the class of the test-

image pixels

Learning

Test

[Moser 10]


18/2418

PDF SAR amplitude ofeach class and each

channel

Copulas

MRF parameter

estimation


find the class of the test-

image pixels

Learning

Test

[Moser 10]

21

2211

*

2211

)(),(

)().()( yy

yFyFC

ypypypmmmmm

mmmmmm


19/24

19

Water Urban Vegetation Overall

DSEM-MRF 99.14 % 98.88 % 84.65 % 94.22 %

K-NN-MRF 96.72 % 96.09 % 99.92 % 97.58 %

CoDSEM (Semivar.) 98.37 % 98.91 % 100 % 99.09 %

CoDSEM (GLCM) 98.62 % 98.42 % 100 % 99.01 %

DSEM-MRFCoDSEM (GLCM)

COSMO-SkymedimageofCavallermaggiore

(Italy)

(ISA,2008)


20/24

20


DSEM-MRF 92.95 % 98.32 % 81.33 % 90.87 %

K-NN-MRF 90.56 % 98.49 % 94.99 % 94.68 %

CoDSEM (GLCM) 91.28 % 98.82 % 93.53 % 94.54 %

CoDSEM (GLCM) K-NN-MRF

TerraSAR-X image of Rosenheim

(Germany) (Infoterra, 2008)


21/24

21


DSEM-MRF 99.81 % 99.03 % 99.99 % 99.61 %

CoDSEM (GLCM) 98.66 % 99.56 % 99.27 % 99.16 %

CoDSEM (GLCM)

COSMO-Skymedimage of Port-au-Prince

(Haiti) (ISA, 2009)

DSEM-MRF


22/24

22

Algorithm validated in the application:

urban/land/water separation on several single-pol. SAR

images

Smoothing effects at spatial borders

Possible improvements:

More sophisticated texture-extraction techniques

Taking into account urban geometry via hierarchical/multiscaleMRFs


23/24

23

French defense agency (DGA)

Italian Space Agency (ISA)

Infoterra (Astrium Services)

Dr. G. Moser and Prof. S. Serpico from Univ. of Genoa

for their helpful comments and fruitful collaboration


24/24

24

[Oliver 04] Oliver, C. and Quegan, S., [Understanding Synthetic Aperture Radar images], SciTech Publishing (2004).

[Jacob 02] Jacob, A. M., Hemmerly, E. M., and Fernandes, D., SAR image classification using a neural classifier based on Fisher criterion, in

[Medical Imaging: Image Processing], Proc. of the VII Brasilian Symposium on Neural Networks (SBRN), 168172 (2002).

[Yang 09] Yang, W., Dai, D., Triggs, B., and Xia, G.-S., Semantic labeling of SAR images with hierarchical Markov aspect models, HAL Research

Report , hal00433600 (2009).

[Cheney 09] Cheney, M. and Borden, B., [Fundamentals of radar imaging], Phladelphia: Society for industrial and applied mathematics (2009).

[Moser 10] Moser, G., Krylov, V., Serpico, S. B., and Zerubia, J., High resolution SAR image classification by Markov random fields and finite

mixtures, Proc. of SPIE7533, 753308 (2010).

[Moser 06] Moser, G., Serpico, S. B., and Zerubia, J., Dictionary-based Stochastic Expectation Maximization for SAR amplitude probability

density function estimation, IEEE Trans. Geosci. Remote Sens. 44(1), 188199 (2006).

[Krylov 09] Krylov, V., Moser, G., Serpico, S. B., and Zerubia, J., Dictionary-based probability density function estimation for high-resolutionSAR data, Proc. of SPIE7246, 72460S (2009).

[Celeux 95] Celeux, G., Cheveau, D., and Diebolt, J., On stochastic versions of the EM algorithm, INRIA Research Report2514 (1995).

[Tison 04] Tison, C., Nicolas, J.-M., Tupin, F., and Maitre, H., A new statistical model for Markovian classification of urban areas in high-

resolution SAR images, IEEE Trans. Geosci. Remote Sens. 42(10), 20462057 (2004).

[Sneddon 72] Sneddon, I., [The use of integral transforms], McGraw-Hill, New York (1972).

[Fjortoft 03] Fjortoft, R., Delignon, Y., Pieczynski, W., Sigelle, M., and Tupin, F., Unsupervised classification of radar images using hidden

Markov chains and hidden Markov random fields, IEEE Trans. Geosci. Remote Sens. 41(3), 675686 (2003).

[Besag 86] Besag, J., On the statistical analysis of dirty pictures,Journal of the Royal Statistical Society48, 259

302 (1986).[Dubes 89] Dubes, R. C. and Jain, A. K., Random field models in image analysis,Journal of Applied Statistics 16(2), 131164 (1989).

[Besag 74] Besag, J., Spatial interaction and the statistical analysis of lattice systems,Journal of the Royal Stat. Soc. 36(2), 192236 (1974).

[Geman 84] Geman, S. and Geman, D., Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images, IEEE Patt. Anal.

Mach. Intell. 6(6), 721741 (1984).

*Chen 04+ Chen, Q. and Gong, P., Automatic variogram parameter extraction for textural classification of the panchromatic IKONOS imagery,

IEEE Trans. Geosci. Remote Sens. 42(5), 11061115 (2004).

[Haralick 73] Haralick, R. M., Shanmugam, K., and Dinstein, I., Textural features for image classification, IEEE Trans. on Systems, Mans and

Cybern. 3(6), 610

621 (1973).

Classification of high resolution sar images using textural features

Documents

Transcript of Classification of high resolution sar images using textural features