In the first half of 2009 we carried out an extensive review of the face recognition literature seeking out any published work that explicitly linked a factor of common interest, from resolution to age, to algorithm performance. Then we analyzed the results, putting them as best we could into a common reference frame. The results is a conference paper presented at
A Meta-Analysis of Face Recognition Covariates, Yui Man Lui, David Bolme, Bruce A. Draper, J. Ross Beveridge, Geoff Givens and P. Jonathon Phillips, IEEE Third International Conference on Biometrics: Theory, Applications and Systems, September 2009.
The paper summarizes findings of how age, gender, race, expression, time between images and image resolution have been shown to influence face recognition performance in past studies. To highlight just one conclusion, while older studies suggested increasing face size beyond a modest size did little to improve recognition performance, more recent studies involving more modern algorithms are starting to suggest that higher resolution face images are useful. The version of the paper published for the conference includes concise summary information needed to see quickly the major findings. The extended version available above includes supplemental material that will be of use to anyone wishing to see additional detail about each of the summary results presented in the main paper.
