6.7 Conclusions
In this chapter, we have provided an overview of satellite SAR image analysis
techniques for sea ice mapping. Based on the characteristics of SAR sea ice
imagery, we have presented a Bayesian method for fast and accurate segmentation
of SAR sea ice imagery. The proposed segmentation scheme is capable of accounting for the spatial correlation effect on both pixel observations and the pixel labels.
The proposed KPCA technique was performed on the image patches to extract
compact and discriminative texture features with Gaussian-like noise characteristics. These KPCA texture features are totally data-driven and capable of revealing
between-class variations. In the proposed Bayesian method, the combined use of
KPCA feature likelihood and the MRF label prior constitutes a coherent and
powerful scheme for automated segmentation of SAR sea ice imagery. Both
simulated SAR images and RADARSAT-2 sea ice images were used for comparing
our segmentation scheme with several other popular methods, such as K-Means,
Gamma mixture, GLCM and MRF. The results evaluated by both visual interpretation and quantitative measures suggested that the proposed method achieved
higher accuracy than the referenced techniques. Moreover, the proposed method
achieved very high time-efficiency, thus may better support the operational segmentation of SAR sea ice imagery.
References
Bagon S (2006) Matlab wrapper for graph cut, Dec [online] Available http://www.wisdom.
weizmann.ac.il/~bagon
Baraldi A, Parmiggiani F (1995) An investigation of the textural characteristics associated with
gray level co-occurrence matrix statistical parameters. IEEE Trans Geosci Remote Sens
33(2):293–304
Barber DG, Ledrew EF (1991) SAR sea ice discrimination using texture statistics: a multivariate
approach. Photogramm Eng Remote Sens 57(4):385–395
Besag J (1986) On the statistical analysis of dirty pictures. J R Stat Soc Ser B 48:259–302
Boykov Y, Veksler O, Zabih R (2001) Fast approximate energy minimization via graph cuts. IEEE
Trans Pattern Anal Mach Intellegence 20(11):1222–1239
Burns BA, Kasischke ES, Nuesch DR (1982) Extraction of texture information from SAR data:
application to ice and geological mapping. International Symposium on Remote Sensing of
Environment, Fort Worth, TX, 6–10 Dec, pp 861–868
Carsey FD (ed) (2013) Microwave remote sensing of sea ice. Online ISBN: 9781118663950,
Geophysical Monograph Series, Wiley. doi:10.1029/GM068
Clausi DA (2001) Comparison and fusion of co‐occurrence, gabor and MRF texture features for
classification of SAR sea‐ice imagery. Atmos-Oceans 39(3):183
Clausi DA (2002) An analysis of cooccurrence texture statistics as a function of grey level
quantization. Can J Remote Sens 28(1):45–62
Clausi D, Yue B (2004) Comparing cooccurrence probabilities and markov random fields for
texture analysis of SAR sea ice imagery. IEEE Trans Geosci Remote Sens 42(1):215–228
134
L. Xu and J. Li
In this chapter, we have provided an overview of satellite SAR image analysis
techniques for sea ice mapping. Based on the characteristics of SAR sea ice
imagery, we have presented a Bayesian method for fast and accurate segmentation
of SAR sea ice imagery. The proposed segmentation scheme is capable of accounting for the spatial correlation effect on both pixel observations and the pixel labels.
The proposed KPCA technique was performed on the image patches to extract
compact and discriminative texture features with Gaussian-like noise characteristics. These KPCA texture features are totally data-driven and capable of revealing
between-class variations. In the proposed Bayesian method, the combined use of
KPCA feature likelihood and the MRF label prior constitutes a coherent and
powerful scheme for automated segmentation of SAR sea ice imagery. Both
simulated SAR images and RADARSAT-2 sea ice images were used for comparing
our segmentation scheme with several other popular methods, such as K-Means,
Gamma mixture, GLCM and MRF. The results evaluated by both visual interpretation and quantitative measures suggested that the proposed method achieved
higher accuracy than the referenced techniques. Moreover, the proposed method
achieved very high time-efficiency, thus may better support the operational segmentation of SAR sea ice imagery.
References
Bagon S (2006) Matlab wrapper for graph cut, Dec [online] Available http://www.wisdom.
weizmann.ac.il/~bagon
Baraldi A, Parmiggiani F (1995) An investigation of the textural characteristics associated with
gray level co-occurrence matrix statistical parameters. IEEE Trans Geosci Remote Sens
33(2):293–304
Barber DG, Ledrew EF (1991) SAR sea ice discrimination using texture statistics: a multivariate
approach. Photogramm Eng Remote Sens 57(4):385–395
Besag J (1986) On the statistical analysis of dirty pictures. J R Stat Soc Ser B 48:259–302
Boykov Y, Veksler O, Zabih R (2001) Fast approximate energy minimization via graph cuts. IEEE
Trans Pattern Anal Mach Intellegence 20(11):1222–1239
Burns BA, Kasischke ES, Nuesch DR (1982) Extraction of texture information from SAR data:
application to ice and geological mapping. International Symposium on Remote Sensing of
Environment, Fort Worth, TX, 6–10 Dec, pp 861–868
Carsey FD (ed) (2013) Microwave remote sensing of sea ice. Online ISBN: 9781118663950,
Geophysical Monograph Series, Wiley. doi:10.1029/GM068
Clausi DA (2001) Comparison and fusion of co‐occurrence, gabor and MRF texture features for
classification of SAR sea‐ice imagery. Atmos-Oceans 39(3):183
Clausi DA (2002) An analysis of cooccurrence texture statistics as a function of grey level
quantization. Can J Remote Sens 28(1):45–62
Clausi D, Yue B (2004) Comparing cooccurrence probabilities and markov random fields for
texture analysis of SAR sea ice imagery. IEEE Trans Geosci Remote Sens 42(1):215–228
134
L. Xu and J. Li