Modeling of target shadows for SAR image classification

Scott Papson, Ram Narayanan

Research output: Chapter in Book/Report/Conference proceedingConference contribution

14 Scopus citations


A recent thrust of non-cooperative target recognition (NCTR) using synthetic aperture radar (SAR) has been to complement the extraction of scattering centers by incorporating information contained in the target shadow. When classifying targets based on the shadow region alone, it is essential that an image be well clustered into its respective shadow, highlight, and background regions. To obtain the segmentation, the intensity and spatial location of a pixel are modeled as a mixture of Gaussian distributions. Expectation- maximization (EM) is used to obtain the corresponding distributions for the three regions within a given image. Anisotropic smoothing is applied to smooth the input image as well as the posterior probabilities. A representation of the shadow boundary is developed in conjunction with a Hidden Markov Model (HMM) ensemble to obtain target classification. A variety of targets from the MSTAR database are used to test the performance of both the segmentation algorithm and classification structure.

Original languageEnglish (US)
Title of host publication35th Applied Imagery and Pattern Recognition Workshop, AIPR 2006
PublisherInstitute of Electrical and Electronics Engineers Inc.
Number of pages1
ISBN (Print)0769527396, 9780769527390
StatePublished - 2006
Event35th Applied Imagery and Pattern Recognition Workshop, AIPR 2006 - Washington, DC, United States
Duration: Oct 11 2006Oct 13 2006

Publication series

NameProceedings - Applied Imagery Pattern Recognition Workshop
ISSN (Print)1550-5219


Other35th Applied Imagery and Pattern Recognition Workshop, AIPR 2006
Country/TerritoryUnited States
CityWashington, DC

All Science Journal Classification (ASJC) codes

  • General Engineering


Dive into the research topics of 'Modeling of target shadows for SAR image classification'. Together they form a unique fingerprint.

Cite this