Using facial attractiveness as a soft biometric trait to enhance face recognition performance

Alnamnakani, Moneera Habeeb, Mahmoodi, Sasan and Nixon, M.S. (2023) Using facial attractiveness as a soft biometric trait to enhance face recognition performance. In, Bourlai, Thirimachos (ed.) Face Recognition Across the Imaging Spectrum. Netherland. Springer Nature. (In Press)

Abstract

Soft biometrics are characterized as a set of traits or features that convey information about an individual, but they cannot be used to individually authenticate a subject due to the fact that they lack distinctiveness and permanence. While soft biometrics capture non-intrusive and less specific traitsof an individual, they can enhance and positively impact the performance ofhuman-based identification systems, including face recognition. To optimize face recognition using soft biometrics, facial features can be derived manually (human labelling), automatically, or semi-automatically (features extracted automatically and refined by an operator). Several features are typically associated with soft biometrics, including age, gender, ethnicity, eye color, and weight, and for facial recognition, these can also include smiling, expression, wearing glasses, and having a facial scar, mark, or tattoo. One facial feature not yet considered for identification is attractiveness based on facial characteristics. Attractiveness has been found to have strength comparable to gender in aiding recognition.This work explores and describes the relationship between attractiveness and beauty and its implications in recognition, psychology, philosophy, and automated analysis. It is surprising that attractiveness has yet to be considered, as many user features are directly related to the human face rather than human perception.Previous computer vision approaches focusing on attractiveness are not well formulated for recognition purposes. In this chapter, we introduce a novel approach that incorporates attractiveness as a facial feature. This feature is derived by comparing faces and then ranking comparisons, making it descriptive in a biometric sense, and it is demonstrated that attractiveness can indeed aid the recognition process. Specifically, in this work, we show that utilizing facial attractiveness improves Face Recognition (FR) performance by more than 3%when used as a soft biometric, and by 4% when used in attractiveness-guided automatic recognition, both tested on the LFW face dataset. These results signify substantial improvements achieved by incorporating attractiveness in two different face recognition frameworks, using standard baseline approaches rather than deep learning to fully establish the fundamental nature of this newly proposed feature.

This record has no associated files available for download.

Contributors

Download statistics