Abstract:
"
Mostly developed countries use Facial Recognition to fight terrorism and to keep their
country safe from various incidents like bomb blasts or crimes. So, in those countries
facial recognition is widely used by security forces like the military and police.
Likewise, most of the private and government institutions and companies also use facial
recognition to identify their employees and for security purposes.
One of the problems of facial recognition is identifying false faces as positives. This
problem is happening throughout the ethnic minorities according to some studies. This
problem justifies that there is ethnic bias in facial recognition systems and algorithms
which are being used currently. This dissertation is a result of the project which aims
to explore the ethnicity bias issue on the current state of the art facial recognition
algorithm Facenet and to provide a novel image selection method to minimize the effect
from the ethnicity bias without making changes to the neural network architecture
This approach modifies the existing facenet approach while classifying the ethnicities
of images of the dataset to minimize the ethnic bias of the facenet. This approach
considers the ethnic distribution of the given data set and uses that to create the small
batches of face images which will be used to train the model. The model trained with
the proposed method has shown significant improvement when considering the AUC
metric which reduces the chance to misclassify a person with underrepresented
ethnicity. Thus, reducing the false acceptance rate which increases the reliability of the
proposed model as well as minority representation. The clustering evaluation tasks
revealed that the model trained with the proposed approach can form near-perfect
clusters while the base model struggles to do so. The solution is available as multiple
jupyter notebooks which will run on google colab."
