Scene Analysis Using Recursive Frequency Domain Correlation with Energy Normalization
Abstract
This thesis describes a scene analysis algorithm which locates targets in a noisy background. Preprocessing is used to detect the edges of objects in a digitized scene. The edges extracted from the scene is then cross- correlated with a template of the target to be found. Cross-correlation is done by using complex-conjugate multiplication in the frequency domain. Areas of high correlation are recorrelated using smaller images which can be processed faster. The potential targets are energy normalized with respect to the template in order to eliminate false correlation with noise. Even in a scene with high energy noise, the algorithms works very well when operated under the constraints of size, orientation, and perspective.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1984
- Accession Number
- ADA151777
Entities
People
- Richard L. Mills
Organizations
- Air Force Institute of Technology