Comparison of Outputs for Variable Combinations Used in Cluster Analysis on Polarmetric Imagery
Abstract
Polarimetric imaging provides potential for highlighting man-made objects amongst complex natural backgrounds because man-made objects emit radiation with a higher degree of polarization than natural environments. More specifically, two techniques, Cluster Analysis (CA) and Principle Component Analysis (PCA) can be combined to process Stoke s imagery by distinguishing between pixels, and producing groups of pixels with similar characteristics. In this study, an algorithm which performs PCA and CA on three to five of the Stoke s imagery at a time was run on the same image subsection for all sixteen possible combinations in order to observe the differences between the combinations. After the data was compiled, the most basic cluster image and corresponding data was compared across all combinations. It was found that the majority of the groups had significantly different mean values at the 95% confidence level, and of this majority, most remained significant at the 99.9% confidence level. In addition, 14/16 of the data sets had a significant proportion of pixels in the smaller cluster group at the 95% confidence level, with 7/14 remaining significant at the 99.9% confidence level.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2008
- Accession Number
- ADA476686
Entities
People
- Melinda Petre
Organizations
- United States Army Research Laboratory