Accession Number:

AD1066095

Title:

A Facial Recognition Algorithm Comparison: Using a Hybrid EigenFace ARTMAP Neural Network vs. the Tracking-Learning-Detection (TLD) Algorithm

Descriptive Note:

Technical Report,01 Jan 2014,01 Feb 2019

Corporate Author:

US Army Research Laboratory Aberdeen Proving Ground United States

Personal Author(s):

Report Date:

2019-01-22

Pagination or Media Count:

18.0

Abstract:

This report describes a comparison of two facial recognition processes for continuous learning. One process used an ARTMAP neural network with features extracted using a modified EigenFace implementation. This was compared with training the Tracking-Learning-Detection TLD algorithm using faces from a television episode. Results indicated that the TLD algorithm was superior to the Hybrid EigenFaceARTMAP EA for the entire episode but that the Hybrid EA algorithm was better for the second half of the episode. The ARTMAP was chosen because it can adaptively train to new vectors without suffering from catastrophic forgetting. However, the TLD algorithm was capable of better online learning and overall performance.

Subject Categories:

  • Numerical Mathematics
  • Cybernetics

Distribution Statement:

APPROVED FOR PUBLIC RELEASE