Automated Segmentation and Extraction of Area Terrain Features from Radar Imagery

Abstract

An automated method for segmenting and extracting certain area terrain features from Synthetic Aperture Radar (SAR) imagery is presented. First, the input radar image is edge-enhanced by passing it through a Sobel operator in order to obtain the required edges for further processing. The unwanted noise, both from the original image source and from the edge operation, is reduced with a low-pass filter. The next step is a region growing process in which pixels of similar gray values in the filtered image are grouped and merged together. A method of selecting an optimum threshold that is essential for region growing is described. The pixels in the image after the region growing operation are further grouped into exactly four different categories, each with its own gray value. The four categories of pixels are finally classified as water, fields, forests, or urban areas depending upon their gray values. A texture measurement scheme and a Bayes classifier are also incorporated into this effort for verifying the classification results. Keywords: Algorithms; Segmentation.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 1990
Accession Number
ADA221096

Entities

People

  • Pi-fuay Chen
  • Richard A. Hevenor

Organizations

  • Geospatial Research Laboratory

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Boundaries
  • C Programming Language
  • Computer Languages
  • Computer Programming
  • Computer Vision
  • Computers
  • Filters
  • Filtration
  • High Resolution
  • Image Processing
  • Image Segmentation
  • Images
  • Lisp Programming Language
  • Low Pass Filters
  • Machine Learning
  • Radar
  • Synthetic Aperture Radar

Readers

  • Computer Vision.