You must have a javascript-enabled browser and javacript and stylesheets must be enabled to use some of the functions on this site.
 Earthnet Home  
Sessions and Session Summaries
First Annoucement
Scientific Committee
ESA Specific Links
Organising Committee
Round Table Discussion Questions
Conference Photos
List of Participants
All papers


A comparison of statistical segmentation techniques for multifrequency polarimetric SAR: region growing versus simulated annealing

Tiziana Macrž Pellizzeri (1), Professor Pierfrancesco Lombardo(1) , Ian Mc Connell(2) , Marco Meloni(1) , Dr. Christopher J. Oliver(2) , and Massimo Sciotti(1)

(1) University of Rome 'La Sapienza', Via Eudossiana 18, 00184 - Rome, Italy
(2) NASoftware, Liverpool, Liverpool, United Kingdom


Two polarimetric segmentation techniques have been devised and implemented by the authors, starting from the generalised maximum likelihood approach with a Wishart distribution: a region growing approach (POL MUM) and global likelihood approach (POLSEGANN) based on the simulated annealing. Both techniques exploit the properties of the covariance matrix of the data, but they proceed with very different approaches to identify the widest possible homogeneous segments. A number of different sets of simulated images with known segments and polarimetric characteristics is used to compare the segmentation performance of the different techniques. Different measures are introduced to compare the achieved segments, among which the probability of correct classification and the fractional overlap between segments that belong to the same region. The comparative performance of the two techniques on a set of SIRC polarimetric images are also discussed.

Extended summary

The extraction of information from polarimetric SAR images, that carry appreciably more information about the observed scene than the single-channel SAR images, is still an area of active research for the many operational polarimetric SAR that are becoming available in the very near future. In particular, an interesting approach to the extraction of information consists in applying a segmentation to the SAR images before any further operation of classification or feature extraction. In fact, it has been clearly demonstrated (see [11], [1]) that large performance improvement can be achieved by first segmenting the SAR image into regions with homogeneous characteristics, and then classifying the resulting global regions. It is therefore of interest to define optimal segmentation techniques for the polarimetric SAR images. Many previous works were related more with the exploitation of both the reflectivity and the phase relationship between the different polarimetric channels to provide useful indications on the characteristics of the different observed objects, allowing a more accurate classification [1-4]. Many examples of both segmentation and terrain classification using polarimetric SAR data can be easily found in literature, with application to very different monitoring and surveillance applications [5-10]. The theory at the basis of effective approaches for the segmentation of polarimetric SAR images has been proposed by the authors in [12], [13]. These works presented the derivation of the generalised likelihood function of a single segment of polarimetric SAR image. This was shown to be the basic step to obtain both the global generalised likelihood of the whole polarimetric SAR image and the split-merge test between two adjacent regions. Then, the global generalised likelihood was used in connection with a simulated annealing approach to derive a polarimetric segmentation scheme, named POLSEGANN. The aim of this paper is to introduce a region growing technique (POL MUM) for segmenting the polarimetric SAR images, based on the optimal split-merge test. The resulting algorithm is then compared to POLSEGANN to assess their comparative theoretical performance both on simulated data and real SAR data. A full set of cases is studied to generate a complete simulated comparison of the techniques. A specific test pattern is selected containing many regions with different shapes and randomly filled with samples from seven different classes. A number of random images are generate in a Monte Carlo simulation with the same statistical properties in each region. Then both POLSEGANN and POLMUM are fed with the same data and the accuracy of the segmentation is evaluated, by averaging the performance over the set of simulations. Different measures are introduced to compare the achieved segments: (i) the probability of correct classification; (ii) the fractional overlap between segments that belong to the same region. Both measures are based on the a priori knowledge on the generated images, that can be used to evaluate the behaviour of the different techniques. In the first case (i), the average confusion matrix is generated, that allows us to discuss the ability to discriminate between adjacent regions belonging to different classes with similar properties. The average probability of correct classification yields a global measure of segmentation correctness, since a bad segmentation also implies a bad classification. However, this is a global measure and it is not very sensitive to the shape of the regions borders. The second measure is selected to analyse the border estimation accuracy and is based on a differential measure of overlap. Both techniques show to operate effectively against the simulated pattern with a little advantage for the POLSEGANN technique in the extraction of the shape of the borders and dealing with the small regions. Moreover, the comparative performance are evaluated in terms of computational complexity. Finally, the proposed segmentation techniques are applied to a set of polarimetric SIR-C images. The segmentation performance against the real data are compared in terms of correct classification using the available ground truth.


[1] Oliver, C.J., and Quegan, S., "Understanding SAR images", Artech House, NY, 1998.

[2] Van Zyl, J.J., Zebker, H.A., Elachi, C., "Imaging radar polarization signatures: Theory and observations", Radio Science, Vol. 22, 1987, pp. 529-543.

[3] Ulaby, F.T. and Elachi, C., Radar Polarimetry for Geoscience Applications, Artech House, Norwood, 1990.

[4] Zebker, H.A., Van Zyl, J.J., "Imaging radar polarimetry: A review", Proceedings IEEE, Vol. 79, 1991, pp. 1583-1606.

[5] Novak, L. M. and Burl, M. C., "Optimal speckle reduction in polarimetric SAR imagery", IEEE Trans. Aerospace Electronic Systems, Vol. 26, 1990, pp. 293-305.

[6] Lee, J., Grunes, M. R. and Mango, S. A., "Speckle reduction in multipolarisation, multifrequency SAR imagery", IEEE Trans. Geoscience and Remote Sensing, Vol. 29, 1991, pp. 535-544.

[7] Lombardo P., Oliver C.J., "Optimum Detection and Segmentation of oil-slicks using polarimetric SAR data", IEE Proceedings on Radar, Sonar and Navigation, Vol. 147, No. 6, December 2000.

[8] Cloude, S.R., Pottier, E., "A Review of Target Decomposition Theorems in Radar Polarimetry", IEEE Trans. on GRS, Vol. 34, 1996, pp. 498-518.

[9] Cloude, S. R. and Pottier, E., "An entropy based classification scheme for land applications of polarimetric SAR", IEEE Trans. Geoscience and Remote Sensing, Vol. 35, 1997, pp. 68-78.

[10] Lee, J.-S. and Hoppel, K. W., "Principal components transformation of multifrequency polarimetric SAR imagery", IEEE Trans. Geoscience and Remote Sensing, Vol. 30, 1992, pp. 686-696.

[11] T. Macrž Pellizzeri, C. J. Oliver, P. Lombardo, "Segmentation-Based Joint Classification of SAR and Optical Images", in print IEE Proceedings on Radar Sonar and Navigation, 2002.

[12] P. Lombardo, C.J. Oliver, "Optimal Classification of Polarimetric SAR images Using Segmentation", IEEE Radar Conference 2002, Long Beach (CA), April 2002.

[13] P. Lombardo, C.J. Oliver, "Optimal Polarimetric Segmentation for the Classification of Agricultural Areas", EUSAR 2002, Koeln, Germany, June 2002.


Full paper


  Higher level                 Last modified: