Hierarchical Relaxation Methods for Multispectral Pixel Classification as Applied to Target Identification.
Abstract
This report provides insights into the approaches toward image modeling as applied to target detection. The approach is that of examining the energy in prescribed wave-bands which emanate from a target and correlating the emissions. Typically, one might be looking at two or three infrared bands, possibly together with several visual bands. The target is segmented, using both first and second order modeling, into a set of 'interesting components' and these components are correlated so as to enhance the classification process. A Markov-type model is used to provide an a priori assessment of the spatial relationships among critical parts of the target, and a stochastic model using the output of an initial probabilistic labeling is invoked. The tradeoff between this stochastic model and the Markov model is then optimized to yield a best labeling for identification purposes. In an identification of friend or foe (IFF) context, this methodology could be of interest, for it provides the ingredients for such a higher level of understanding. Keywords: Multispectral classification; Wave bands; Split and merge algorithm; Generalized slope facet model; Sobolev model; Kalman filters; Nonlinear filtering; Boundary classification; Moment area method; Relaxation methodology; Stochastic labeling; Quad tree; Interesting components; Background replacement concept.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1985
- Accession Number
- ADA158694
Entities
People
- E. A. Cohen Jr
Organizations
- Naval Ordnance Laboratory