Facial asymmetry versus facial makeup

Variations in facial appearance resulting from the application of ordinary makeup products, such as eyeliner and lipstick, are challenging for automated facial recognition. This work studies the potential of using the measurement of the inherent asymmetry between the two halves of a person's face presented in [18] as a helpful feature to overcome this challenge. We hypothesized that inherent facial asymmetry is not completely concealed by makeup or might even be increased by the application of makeup. To test our hypothesis, we applied the Eigenfaces algorithm to classify the faces of 67 ethnically diverse individuals in our labeled database, with and without makeup, based on the asymmetry feature. The database was designed so that all variations, except the application of makeup, were fixed. The use of Eigenfaces in this preliminary evaluation was intentional; the feature classification process of this method is simple, which makes the effects of asymmetry features on the classification process easier to observe, and Eigenfaces are widely used in popular computer vision packages, such as OpenCV. The results show that recognition accuracy improved by 42.36% when using our asymmetry feature compared to the classical Eigenfaces algorithm. These results are encouraging and will serve as the baseline for future experiments on new datasets with other more robust classifiers.