Abstract:
Taking an advantage of improving technology, dozens of software solutions have developed
which use biometrical features of an individual as a source. The face of a person was always
a target biometrics for computer scientists. Moreover, the research originated from
recognition of a person by their face have a history of over 50 years. Primarily, the statistical
and mathematical approaches were preferred, in fact there were not that many options back
then. By invention of advanced Machine Learning (ML) and Deep Learning (DL) techniques
research flow changed from holistic matching approaches in which whole face is considered
as an input and processed completely to feature localization backed techniques, which are
extracting base features such as eyes, nose, mouth and doing mathematical measurements
to generate specific identifier (embedding) for the faces. Although its short history local
feature extraction methods became more popular, due to their high accuracies. In this
research thesis, one of the main goals is to explore current deep learning-based approaches
and build pipeline by using gained knowledge. By that purpose, different state of art models
is explored for face detection and recognition. At the end, by using combination of some of
these models a pipeline is developed which works sequentially by detecting the faces at first
stage and recognizing the faces later. Majority of these techniques which explored can be
efficiently used in specific implementation areas such as Security, Access control etc.
spheres.