Vision Strategies and ATR Performance: A Mathematical/Statistical Framework and Critique
Abstract
The broad goal of this project is to formulate mathematical models which are sufficiently general to support an analysis of particular algorithms for target detection and recognition from the perspective of classical statistics and information theory. When approached from this viewpoint, questions about the performance of an algorithm for detection, recognition or identification translate into familiar problems in estimation, complexity and hypothesis testing. Consequently, an arsenal of powerful results from statistics and information theory, for example results about optimal codes, most-powerful tests, inference and efficiency, and the complexity of testing highly composite hypotheses, can be exploited to achieve a deeper understanding of the ATR problem. The focus of the research is on performance metrics, various measures of an algorithms performance such as probability of detection, probability of "false alarm," bias/variance tradeoffs for algorithms that learn from training data, and computational complexity.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 03, 2004
- Accession Number
- ADA425478
Entities
People
- Donald E. Mcclure
Organizations
- Brown University