Rochester Institute of Technology
Selection of Optimal Background Estimation Methods for Unstructured Detectors
Abstract
dc:description.abstractTarget detection algorithms determine the presence of a target in a pixel by comparing the spectrum of a given pixel to the both the target and an estimate of the background. Because of this target detection performance is particularly sensitive to the way that the background is characterized. We will investigate the performance of a number of Background Characterization Methods (BCMs) described by J. West by applying the S. Rotman target implantation method. Our goal will be to determine which BCM works best for a particular image and a particular target when using the Generalized Likelihood Ratio Test (GLRT) target detector. Normally, target detection performance is quantified in some image wide metric, such as the Average False Alarm Rate (AFAR). We develop a method that allows us to evaluate target detection performance on a pixel by pixel basis for each of the BCMs. We decompose the unstructured background detector into a number of components that allow us to reconstruct the detection statistics for any unstructured background detector at any arbitrary target concentration. This allows us to calculate the minimum target concentration required to detect a target while maintaining a given Constant False Alarm Rate (CFAR). We can then compare this minimum concentration to determine the optimal background for each pixel. This method has been applied to the Forest Radiance I image collected over the U.S. Army Aberdeen Proving Ground in August 1995, as well as the Desert Radiance II image collected over the U.S. Army's Yuma Proving Ground in Arizona in June 1995.
Degree
thesis:*- Name thesis:degree_name
- Imaging Science (MS)
- Level thesis:degree_level
- Thesis
- Discipline thesis:degree_discipline
- Chester F. Carlson Center for Imaging Science (COS)
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Ferdinandus, Manuel R.
- Contributors dc:contributor
-
- John R. Schott
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.rit.edu/theses/8220