Scene Analysis: Non-Linear Spatial Filtering for Automatic Target Detection.
Abstract
This work focuses on a method for two-dimensional pattern recognition. The method includes a global search scheme for candidate windows of interest, based on Fourier domain cross-correlation. A method to normalize the input scene by local rectangular regions, in an attempt to efficiently approximate search windown mormalization, is presented. Also developed is a candidate window (potential target) similarity measure, based on the normalized L1 and Euclidean distances, which is independent of the template DC value and its energy. Observations on the performance of the algorithm applied to visual spectrum photographs of tanks in a realistic environment are included. Also included is the software to implement the algorithm on a Data General Eclipse S/250 minicomputer. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1982
- Accession Number
- ADA124788
Entities
People
- James H. Cromer
Organizations
- Air Force Institute of Technology