See News for 2010 for information about our more recent OpenSource releases.
Algorithms - The CSU Face Identification Evaluation System, Version 5.1
On July 23, 2010 we released version 5.1 of our software. This is a minor refinement of the 5.0 release, see the notes below for details.
On May 1, 2003 we released version 5.0 of our software. This software includes simple C source code for four distinct face recognition algorithms along with image pre-processing software, statistical analysis software and scripts to run experiments comparable to the original FERET evaluation and some of our own more recent experiments.
This release includes a Users Guide. The Users Guide is now the definitive source for information on the algorithms available and the other tools provided in the CSU Face Identification Evaluation System.
The algorithms provided in the CSU Face Identification Evaluation System are all written in ANSII C and are very portable. However, significant use is made of Unix scripting to assemble component algorithms into end-to-end experiments, and users will be able to make much quicker and direct use of the system if running under some variant of Unix.
Changes since version 5.0 are:
- The way wavelet responses are normalized has been changed in the EBGM algorithm both to improve performance and be more consistent with the original work by USC.
Significant changes since version 4.0 are summarized in the Users Guide. Briefly,
- A full implementation of the Elastic Bunch Graph Matching (EBGM) algorithm is now included.
- The Bayesian Intrapersonal/Extrapersonal Classifier (BIC) code has been rewritten and made simpler to use.
- The permutation analysis tools now computes sample distributions for differences of recognition rates.
- Distance measures based upon Mahalinobis transformations of image data have been improved and clarified.
- New simpler Unix scripts are included. These will carry out a full experiment using all four face recognition algorithms and produce as output a web-based summary of the results that looks like this.
We recommend those who are using earlier versions of our software upgrade to this release.