CSU Face Recognition Resources

CSU Baseline Algorithms 2011, January 2012 Release


Working with NIST, we have created a two OpenSource baseline face recognition algorithms. The first is a local region PCA algorithm (LRPCA) first release in 2010. The second is an LDA based algorithm released in October 2011.

Here is the download link:

A README file comes with the distribution and is available here for quick inspection.

Brief Overview

While these algorithm will be useful on a variety of datasets, they have been developed specifically to provide a relatively strong lower bound for performance on the Good, the Bad and the Ugly (GBU) Challenge Problem. The paper An Introduction to the Good, the Bad, and the Ugly Face Recognition Challenge Problem is a good place to start to better understand the challenge problem as LRPCA Baseline Algorithm.

We ask that anyone downloading this software fill out the form which appears when using the download link below. It helps us greatly if we can know who is using this code and why. We are also appreciative of any information you might pass back to us regarding results you obtain with the algorithm.

The algorithms are written in a combination of Python and C++ and they require both OpenCV as well as the Scientific Tools for Python and Python Image Processing libraries. The code is available both as source only as well as bundled into a ready to run Virtual machine.

History of release updates:

August 7, 2012
This release is a minor update to change the naming convention for the LDA algorithm which, moving forward, will be described as the Cohort LDA algorithm. Use of the earlier name, LDAIR, is discouraged as it led to confusion relative to IR meaning Infrared. Functionality this release will behave the same as the January 2012 release.
January 27, 2012
The October 26, 2011 release is made official with the dropping of the adjective 'beta'; the code is the same. The version supplied with a Virtual machine is changed to run with a Virtual Box VM in place of a Parallels VM.
October 4, 2011
Beta Release of new code package including both the 2010 LRPCA algorithm as well as the new LDA baseline algorithm. Option to download a complete ready-to-run Virtual Machine added.
March 23, 2011
Minor update to correct flaw in similarity matrix reading. Most users working only with this code should not see a difference, but the fix is important for those working with BEE (Biometric Evaluation Environment) codes written by others.
August 16, 2010
Includes numerous internal changes to allow execution to continue even when there problems with the data such as corrupted face image files, incorrect eye coordinates, etc. The actual recognition algorithm did not substantively change and levels of recognition performance should not differ from those seen with the previous verison.
July 28, 2010
This is the first 'pre-release' version of the LRPCA algorithm.

Acknowledgement: This work was supported by the Technical Support Working Group (TSWG) under Task SC-AS-3181C. Communication may be directed to Ross Beveridge at ross at cs dot colostate dot edu

Page last updated August 7, 2012 ... Ross Beveridge


03:38:04 AM (Colorado)
Thursday 25th of August 2016
Contact: Ross Beveridge

Apply to CSU | Contact CSU | Disclaimer | Equal Opportunity
Colorado State University, Fort Collins, CO 80523 USA
© 2010 Colorado State University