Adaptive Model Based ATR System.
Abstract
This project develops a SAR-ATR system that uses invariant histograms and deformable templates. An invariant histogram is a histogram of geometric invariants given by primitive feature sets. Deformable template matching examines the existence of an object by superimposing templates over potential energy fields derived from the image so that it generates the minimum deformation (deformation energy) and the best alignment of the template with features (potential energy). This system has two modes: off-line and on-line. In off-line mode, it generates a library for indexing and deformable templates for verification. In on-line mode, by calculating an invariant histogram from an input image, it performs the deformable templates, it determines the most likely pose and class of the target. We have demonstrated the effectiveness of these two techniques for robust SAR recognition using occluded and camouflaged target images. By analyzing the evaluation results, we have proposed three extensions of the system: dense sampling for robust recognition, partial view windows for robust indexing under occlusion, and photometric invariants for robust verification under camouflage. Some of these ideas have been evaluated; they are quite promising.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1996
- Accession Number
- ADA327472
Entities
People
- Dean Pomeleau
- Katsushi Ikeuchi
- Khotara Ohba
- Robert Collins
- Takeshi Shakunaga
Organizations
- Carnegie Mellon University