Return to search

Computational Face Recognition Using Machine Learning Models

Faces are among the most complex stimuli that the human visual system
processes. Growing commercial interest in face recognition is encouraging, but it
also turns out to be a challenging endeavour. These challenges arise when the
situations are complex and cause varied facial appearance due to e.g., occlusion,
low-resolution, and ageing. The problem of computer-based face recognition
using partial facial data is still largely an unexplored area of research and how
does computer interpret various parts of the face. Another challenge is age
progression and regression, which is considered to be the most revealing topic
for understanding the human face changes during life.
In this research, the various computational face recognition models are
investigated to overcome the challenges posed by ageing and occlusions/partial
faces. For partial face-based face recognition, a pre-trained VGGF model is
employed for feature extraction and then followed by popular classifiers such as
SVMs and Cosine Similarity CS for classification. In this framework, parts of faces
such as eyes, nose, forehead, are used individually for training and testing. The
results showing that there is an improvement in recognition in small parts, such
as recognition rate in forehead enhanced form about 0% to nearly 35%, eyes
from about 22% to approximately 65%. In the second framework, five sub-models
were built based on Convolutional Neural Networks (CNNs) and those models
are named Eyes-CNNs, Nose-CNNs, Mouth-CNNs, Forehead-CNNs, and
combined EyesNose-CNNs. The experimental results illustrate a high recognition
rate when it comes to small parts, for example, eyes increased up to about
90.83% and forehead reached about 44.5%. Furthermore, the challenge of face
ageing is also approached by proposing an age-template based framework,
generating an age-based face template for enhanced face generation and
recognition. The results showing that generated new aged faces are more reliable
comparing with state-of-the-art.

Identiferoai:union.ndltd.org:BRADFORD/oai:bradscholars.brad.ac.uk:10454/19169
Date January 2021
CreatorsElmahmudi, Ali A.M.
ContributorsUgail, Hassan
PublisherUniversity of Bradford, Faculty of Engineering and Informatics
Source SetsBradford Scholars
LanguageEnglish
Detected LanguageEnglish
TypeThesis, doctoral, PhD
Rights<a rel="license" href="http://creativecommons.org/licenses/by-nc-nd/3.0/"><img alt="Creative Commons License" style="border-width:0" src="http://i.creativecommons.org/l/by-nc-nd/3.0/88x31.png" /></a><br />The University of Bradford theses are licenced under a <a rel="license" href="http://creativecommons.org/licenses/by-nc-nd/3.0/">Creative Commons Licence</a>.

Page generated in 0.0021 seconds