The paper explores the use of Independent Component Analysis (ICA) for face recognition, comparing it to Principal Component Analysis (PCA). ICA is a generalization of PCA that is sensitive to higher-order statistics, which are important for capturing complex facial features. The authors performed ICA on face images from the FERET database using two different architectures: one treating images as random variables and pixels as outcomes, and the other treating pixels as random variables and images as outcomes. The first architecture found spatially local basis images, while the second produced a factorial face code. Both ICA representations outperformed PCA in recognizing faces across days and changes in expression. A combined classifier using both ICA representations achieved the best performance. The ICA representations were also found to be more robust to noise and variations in lighting, hair, makeup, and facial expressions. The study suggests that ICA, particularly the factorial code representation, can provide more effective and robust face recognition methods.The paper explores the use of Independent Component Analysis (ICA) for face recognition, comparing it to Principal Component Analysis (PCA). ICA is a generalization of PCA that is sensitive to higher-order statistics, which are important for capturing complex facial features. The authors performed ICA on face images from the FERET database using two different architectures: one treating images as random variables and pixels as outcomes, and the other treating pixels as random variables and images as outcomes. The first architecture found spatially local basis images, while the second produced a factorial face code. Both ICA representations outperformed PCA in recognizing faces across days and changes in expression. A combined classifier using both ICA representations achieved the best performance. The ICA representations were also found to be more robust to noise and variations in lighting, hair, makeup, and facial expressions. The study suggests that ICA, particularly the factorial code representation, can provide more effective and robust face recognition methods.