Region-Of-Interest Correlation Filters.
Abstract
This effort investigated Region-of-Interest (ROI) filter formulations for Binary Phase-Only Filter (BPOF) implementation in coherent optical correlator systems. A Region-of-Interest may be broadly defined as a localized region in a sensor pattern (an image) that has a high probability of containing a pattern (a target image) of interest for an application such as automatic target recognition (ATR). Such filters are useful in the early stages of an ATR process because they can limit the total area of an input image that is processed, thus reducing the required processor throughput (and size, weight, and power consumption). Few if any formal procedures for designing ROI correlation filters have existed prior to this effort, while formulations for more specific filters matched to a narrow range of targets and target aspects are relatively numerous. The work reported here developed and applied concepts of normalized BPOF distance, cluster analysis, and the least aggregate square error (LASE) correlation filter formulation to yield a successful ROI filter formulation procedure, called the cluster-based ROI BPOF. The formulation was implemented in software suitable for use on desktop computers, and the validity of the resulting ROT filters was demonstrated on a realistic test problem.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1996
- Accession Number
- ADA321132
Entities
People
- David L. Flannery
Organizations
- University of Dayton