Feature Evaluation Tools
Abstract
Measurements or transformations of measurements (i.e., features) are often used for distinguishing between items or groups of interest. For example, cepstral features are useful for speaker identification. Given a large set of possible features, the problem is choosing which to use and which not to use in order to optimize speed and performance. A commercially available software package, MATLAB (Mathworks, Inc.) has been used to implement previously developed methods of feature evaluation tools. These tools include the well known Mahalanobis and Battacbaryya distances, along with a class inter-to-intra measurement ratio. A synthetic data generator for creating multi-variate Gaussian classes was also implemented to aid in the development of the evaluation tools.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2002
- Accession Number
- ADA402546
Entities
People
- Brian Costello
Organizations
- Air Force Research Laboratory