ARENA: Memory-based Face Recognition

Researchers: Terence Sim, Rahul Sukthankar, Matthew Mullin, Shumeet Baluja

Abstract

We show that an extremely simple, memory-based technique for view-based frontal face recognition can outperform more sophisticated algorithms that use Principal Components Analysis (PCA) and neural networks. This method does not perform any complex feature extraction, nor does it incorporate any face-specific information. This technique is closely related to correlation templates; however, we show that the use of novel distance metrics greatly improves performance. We show that augmenting the memory base with additional, synthesized face images results in further improvements in performance. Extensive empirical testing on two standard face recognition datasets, the ORL and FERET databases is presented, and direct comparisons with published work show that our algorithm achieves comparable (or superior) results. This paper further demonstrates that our algorithm has good asymptotic computation and storage behavior, and is ideal for incremental training. Our system has been integrated with a neural-network based face detection system into a real-word visitor identification system that has been operating successfully in an outdoor environment with uncontrolled lighting for several months.

Related Publications


Face Recognition Datasets

The ORL dataset (40 people, 10 images each) is available here.

The FERET dataset may be obtained from P. Jonathon Phillips (jonathon@nist.gov). A subset of the FERET images was used, since the experiments described above require multiple training images for each individual. While we cannot redistribute these images, we have provided a complete list of filenames here.


Rahul Sukthankar (rahuls@cs.cmu.edu),