FACE ALIGNMENT AND NORMALIZATION: FOUNDATIONS, TECHNIQUES AND ADVANCES FOR ROBUST FACE RECOGNITION

Face alignment and normalization are the two irreplaceable
foundations of the modern face recognition systems that cannot be
assembled without prerequisite work. The purpose of these
underlying processes is to canonicize the geometric arrangement as
well as photographic look of facial images and to change extensively
diverse, in-the-wild captures into a standard form that is amenable to
dependable feature extraction and matching. This is the extensive
paper that involves the in-depth analysis of the history, modern
trends, and perspectives in the development of face alignment and
normalization techniques. We begin with a clarification of their
significant role in mitigating the adverse effect of the essential
nuisance factors, such as pose, illumination, expression, and occlusion. The paper continues by giving a systematic taxonomy and a detailed
discussion of fundamental methods, including the classical model- based methods such as Active Shape Models, the period of
discriminative cascaded regression, and the modern model of deep
learning-based holistic alignment and feature-level normalization
networks. Much importance is taken on advanced and new issues:
3D-aware alignment to extreme pose correction, adversarial and
disentangled representation learning to achieve photometric
invariance, and the role of equitable normalization to reduce
demographic bias in order to promote algorithmic fairness. More so, we consider how these modules can be strategically integrated into
effective end-to-end recognition pipelines and optimized to execute
on resource-constrained edge devices. This review summarizes the
insights and findings of more than fifty publications that are regarded
as seminal with an aim of formulating how advanced alignment and
normalization have become more central than peripheral, of
facilitating technologies. The development has played a crucial role in
the transformation of face recognition as a limited laboratory use to a
powerful and efficient biometric technology that can be able to
function in the dynamic and challenging real world environment.