Spatial correlation features for SAR images in a small sample size context

  • Roberto Vaccaro
  • Silvana Dellepiane
Poster Session D: Biomedical Applications, Detection, Control & Surveillance, Inspection, Optical Character Recognition
Part of the Lecture Notes in Computer Science book series (LNCS, volume 1311)


The information related with the spatial correlation properties of textured images represents the topic of the present paper. The correlation estimate task is addressed, taking into account the stability problem when small and irregularly shaped training regions are available, as the case of Remote Sensing data of the SAR type.

In these situations, the classical estimate based on large and rectangular training areas shows a large variance and, as a consequence, classification results quality strongly decreases as the training area dimensions decrease.

The proposed approach is based on the simplified assumption of independent and separable spatial correlation properties in the slant and azimuth directions, and it takes advantage of one-dimensional processing to reduce the computation load. Two one-dimensional correlation functions are then easily extracted from small and irregular training areas, and they are successively applied for a classification process, on the basis of a maximum likelihood criterion. Theoretical and experimental comparisons with the classical two-dimensional approach are presented.

Even though some information is lost in the proposed method, larger spatial neighbourhoods can be considered with only a linear increase of computation load. The results achieved on SAR test images show a significant increase in classification accuracy, proving that the simplified one-dimensional approach correctly takes into account spatial information to the end of the classification problem.


Correlation Function Spatial Correlation Training Area Azimuth Direction Training Region 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


  1. [1]
    Athanasios Papoulis, “Probability, Random Variables, and Stochastic Processes”, McGraw-Hill International Editions, 1987.Google Scholar
  2. [2]
    A.Bruce Carlson, “Comunication Systems”, McGraw-Hill International Editions, 1987.Google Scholar
  3. [3]
    F.T.Ulaby, F.Kouyate, B. Brisco,T.H.Lee Williams “Textural information in SAR Images”, IEE transaction on Geoscience and Remote Sensing, Vol. GE-24, No 2 March 1986Google Scholar
  4. [4]
    J.P Hoffbeck,D.A. Landgrebe, “Covariance Estimation for Classifing High Dimensional Data”, 0-7803-2567-2,IEEE, 1995.Google Scholar
  5. [5]
    E. Rignot, R. Chellappa,“Segmentation of Synthetic-aperture-radar complex data”, J.Opt.Soc.Am. A/Vol.8,No. 9/September 1991.Google Scholar
  6. [6]
    A.V.Oppenheim,R.W.Shafer,“Elaborazione numerica dei segnali”, Franco Angeli, 1990.Google Scholar
  7. [7]
    MAESTRO I, Airborne Synthetic Aperture Radar (SAR) Campaign, “DATA REPORT”, European Space Agency, 1990.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1997

Authors and Affiliations

  • Roberto Vaccaro
    • 1
  • Silvana Dellepiane
    • 1
  1. 1.Department of Biophysical and Electronic EngineeringUniversity of GenoaGenoaItaly

Personalised recommendations