The statistical learning methods in image processing and facial recognition
[ X ]
Tarih
2021
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Altınbaş Üniversitesi
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
The aim of this thesis is to develop a new approach for The Statistical Learning Methods in
image processing and Facial Recognition using the deep learning techniques in python. In the
recent years there have been significant advances in face recognition by using deep neural
networks. One of the potential next steps is to develop optimized 3D facial recognition. Shifting
from 2D to 3D increases complexity of the problem by adding an- other dimension to data,
making possible solutions more resource hungry. In this thesis we investigate different depth
camera based facial recognition techniques and test their performance by deploying them on an
embedded processor. We focus on applications for embedded systems and use a small lowresolution time of flight (ToF) camera with our system to keep overall system portable and
compact. All faces images are then projected on the feature space (“face space”) to find the
corresponding coordinators. The face space is composed of “Eigenfaces” or “Fisherfaces” which
are actually eigenvectors found after doing a matrix composition - Eigen decomposition. At the
heart of Eigenface method is the Principal Component Analysis (PCA) - one of the most popular
unsupervised learning algorithms - while Fisherface is a better version of the previous one which
makes use of both Principal Component Analysis and Linear Discrimination Analysis (LDA) to
get more reliable results. Both methods would be examined deeply in their working principles as
well as their potential applications in reality before coming to conclusion about the advantage and drawback of each one. In order to evaluate the performance of the developed techniques, we
will create a dataset of 10 identities captured with a low-resolution depth camera and use it for
both training and testing. The algorithms were realized by Python in Anaconda. Given initial
images in the database, the program can detect and recognize the human faces in the provided
pictures before saving them in the database to improve the calculation accuracy in the future.
After evaluation, the recognition general results are exported on the screen with details included
in the text files.
Açıklama
Anahtar Kelimeler
Machine Learning, Facial Recognition, Low Computational Power, Linear Discrimination Analysis (LDA)
Kaynak
WoS Q Değeri
Scopus Q Değeri
Cilt
Sayı
Künye
Rafeeq, S. O. (2021). The statistical learning methods in image processing and facial recognition (Yayınlanmış yüksek lisans tezi). Altınbaş Üniversitesi, Lisansüstü Eğitim Enstitüsü, İstanbul.
