Image Metrics
Abstract
Most Automatic Target Recognizer (ATR) development and testing has been conducted in the community without a general means for objectively and quantitatively evaluating performance. It is clear that the performance of any ATR is dependent upon the quality of the input singal. For example, as the noise or clutter in an image becomes increasingly severe, the ATR performance will correspondingly degrade. Therefore it is important to have an objective and quantitative measure of image signal quality by which ATR performance may be gauged. The objectives of this work are to provide and validate a means for quantitative measurement of image signal quality, for use in evaluation of ATR's. In this work we address the ATR function of detection, as performed by a front-end region of interest operator or interest point locator. We also treat both targets and clutter in order to address both true detection rates and false alarm rates. This work is to provide a limited analysis of existing image metrics. Target-to-interference ratio (TIR) is the principal metric. Alternate forms of TIR are also proposed and analyzed. In addition, a limited validation exercise is performed for the TIR metrics, as well as for metrics developed under an Air Force contract called Criteria for Target Recognizer Evaluation (CTRE). The validation exercise consists of determining the degree of correlation between image metrics and the Haar region of interest operator. Two versions of the region of interest operator are used; both are based on a Haar detector that is used in the Army Automatic Target Cuer (ATC) ATR.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1986
- Accession Number
- ADA192213
Entities
People
- A. B. Lucero
- G. B. Silverman
- J. W. Bair
- R. R. Ramroth