A statistical assessment of whitened-PCA/SVD under variable environmental constraints

This work is directed to evaluating Whitened Principal Component Analysis and Singular Value decomposition (PCA/SVD) face recognition algorithm under variable facial expression. The proposed template-based algorithm is tested on some created face database captured along the universally accepted principal emotions. Their recognition distance from the neutral pose are recorded in multivariate sets and prepared for statistical evaluation. The repeated measures design (multivariate method) is used to test for significance difference in the study constraints when being recognized by the propose recognition algorithm. The entire recognition processes and statistical evaluation were modeled using GNU Octave. After experimental runs, recognition results showed that, Whitened PCA/SVD algorithm has an encouraging recognition performance of recognizing images under various principal expressions. The statistical evaluation revealed that, significant difference existed in average Euclidean distance (recognition distance) of the study expressions. Specifically, significant difference existed between the average recognition distance for the constraints (Happy vs Surprise and Sad vs Surprise). All other constraints (facial expressions) considered have pairwise insignificant difference in their recognition distances.