Computational and performance aspects of PCA-based face-recognition algorithms

Algorithms based on principal component analysis (PCA) form the basis of nu merous studies in the psychological and algorithmic face-recognition litera ture. PCA is a statistical technique and its incorporation into a face-reco gnition algorithm requires numerous design decisions. We explicitly state t he design decisions by introducing a generic modular PCA-algorithm. This al lows us to investigate these decisions, including those not documented in t he literature. We experimented with different implementations of each modul e, and evaluated the different implementations using the September 1996 FER ET evaluation protocol (the de facto standard for evaluating face-recogniti on algorithms). We experimented with (i) changing the illumination normaliz ation procedure; (ii) studying effects on algorithm performance of compress ing images with JPEG and wavelet compression algorithms; (iii) varying the number of eigenvectors in the representation: and (iv) changing the similar ity measure in the classification process. We performed two experiments. In the first experiment, we obtained performance results on the standard Sept ember 1996 FERET large-gallery image sets. In the second experiment, we exa mined the variability in algorithm performance on different sets of facial images. The study was performed on 100 randomly generated image sets (galle ries) of the same size. Our two most significant results are (i) changing t he similarity measure produced the greatest change in performance, and (ii) that difference in performance of +/- 10% is needed to distinguish between algorithms.