Submitted By: Swaran Singh

Date: 20/03/2024
 Facial Recognition
Face features are used by a biometric technology known as face recognition to identify or
authenticate a person. It looks at and recognises specific patterns in a person's face, such
as where their lips, nose, eyes, and other facial characteristics are positioned, by using
computer algorithms. These algorithms use distinctive traits from a person's face
photographs to identify them by comparing them to a database of faces that have been
identified

1.Common Applications of Facial Recognition:

 Security and Access Control: Facial corporations and educational
recognition is a technology that is institutions to monitor employee
frequently included in security systems attendance. Because it recognises and
to allow access to restricted areas or records people's presence automatically,
protected equipment. It might provide a it streamlines the processes involved in
maintaining attendance.
more useful and secure means of
 Personal Devices: A lot of laptops,
authentication than conventional
tablets, and smartphones now come
techniques like keycards or passwords. equipped with facial recognition, which
. is a convenient way to unlock devices
 Law enforcement and Surveillance: and access private information. It offers
To locate individuals of interest or an additional layer of security over
identify suspects in public spaces, law traditional passcodes or fingerprint
enforcement organisations employ recognition.
facial recognition software for  Social media and Photo Tagging:
surveillance. It can enhance public Facial recognition technology is widely
safety and aid in criminal used by social networking firms to
investigations. automatically tag photos, making it
. easier for users to identify and share
 Attendance Tracking: Facial pictures of themselves and their
recognition technology is used by both acquaintances

2. Literature Review: Algorithms for Machine

Learning in Identifying Faces
Over the years, several academics have proposed a wide range of machine learning methods. The
report's next section will discuss the most popular and extensively utilised face recognition
algorithms.
 Convolutional Neural Networks (CNNs): Because of their ability to automatically extract
hierarchical characteristics from the target face, CNNs have become the state-of-the-art
technology for facial identification. In their groundbreaking research on handwritten digit
recognition in the 1990s, LeCun et al. introduced CNNs [LeCun et al., 1998]. Since then, a
wide range of visual identification tasks, including facial recognition, have been widely
implemented using CNNs. By winning the 2012 Image Net Large Scale Visual Recognition
Challenge (ILSVRC). showed off the capabilities of CNNs. In facial recognition, CNNs are
quite good at extracting discriminative features from raw pixel data, therefore they do not
need to be manually generated for feature extraction.

 Support Vector Machines (SVM): SVMs are widely utilised in facial recognition
applications because of their ability to learn judgement boundaries between different
identities or visual traits. Since their first introduction by Vatnik et al. [Vatnik, 1995],
support vector machines (SVMs) have been used for a variety of classification tasks,
including facial recognition. Finding the hyper plane in the feature space that most
effectively divides different classes is how SVMs work. Alongside manually constructed
features or feature vectors that were derived using techniques like PCA or LDA, SVMs
have been used in facial recognition applications.

 PCA, or principal component analysis: Principal Component Analysis is a

dimensionality reduction technique that is often used to extract significant characteristics
from facial image data. PCA was first introduced by Pearson [Pearson, 1901] and is used
to identify orthogonal axes, or main components, along which the variation of the data is
greatest. In order to reduce the dimensionality of face photos while retaining the most
amount of information, PCA is employed in facial recognition. Following PCA, the
condensed feature space is fed into a classifier, like SVM or KNN, for identity
classification.
.

 K-Nearest Neighbours (KNN): KNN is a straightforward instance-based learning

technique that is applied to facial recognition and other classification problems. A test
sample is classified using the KNN technique, which Fix and Hodges [Fix and Hodges,
1951] devised, based on the majority vote of the test sample's closest neighbours in the
training dataset. KNN performs facial recognition by comparing the features of a test face
with those of its nearest neighbours in a training dataset. Although KNN is easy to
understand and implement, it may have substantial computational costs, particularly in
large-scale facial recognition systems.

2.Convolutional Neural Networks (CNNs): A

Brief Overview
The biological processes present in the human brain serve as the model for a type of deep
learning algorithms known as convolutional neural networks, or CNNs for short. These
neural networks are an excellent option for tasks like facial and image recognition
because of their exceptional ability to interpret visual data. Recognition: A CNN is made
 up of several layers of neurons, each of which is in charge of extracting particular
features from the input data. Convolutional, pooling, and fully linked layers are the most
common types of these layers. Convolutional layers use filters to find edges, textures, and
other patterns and features in the input image. Afterward, the feature maps are down
sampled by the pooling layers, which lowers the dimensionality of the data. Ultimately,
classification is carried out by the fully linked layers using the features that were
extracted.

3. CNNs for Facial Recognition Technical Specifications

 Data Collection and Preprocessing: A  Feature Extraction: The trained CNN
large dataset of face pictures is can recognise features like glasses or
collected and pre-processed to ensure facial hair, as well as important details
consistency in lighting, posture, and like the mouth, nose, and eyes,
facial expression. Preprocessing automatically from face pictures.
methods include things like resizing
 Classification: Following feature
images, levelling pixel values, and
extraction, CNN categorises the photos
aligning facial landmarks.
to determine the identity of the subject.
 Training the CNN’s: CNNs are trained
using datasets that have already been At this point, the retrieved features are
pre-processed. During training, the compared to those of people who are
network learns to recognise facial known and stored in a database.
features by adjusting the weights of its  Verification and Identification: Facial
neurons using gradient descent and recognition systems can be used for two
back propagation. CNN is constantly purposes: verification (asking if a
improving its ability to accurately person is who they say they are) and
classify faces based on the information identification (comparing an unknown
it collects. face to a known person from a
database).

3.Analogy:
To comprehend the technical explanation of CNN given above, we can apply an analogy.
Consider CNN to be a highly skilled investigator with a broad variety of investigation
skills. Each skill helps the investigator recognise the target's lips, nose, and eyes, among
other face traits.

 Gathering and Preparing Data: First, we gather a lot of face pictures and make sure
they're all uniformly prepared, with good lighting and head alignment. As a source of
data for investigators
 Training of CNN: Afterwards By continuously presenting a vast quantity of images, we
train our investigators (CNNs) to recognise different characteristics. It adjusts its
"thinking" as it learns by taking into account how well it can identify faces.
  Spotting traits: Our investigator has been trained to automatically identify important
face traits, like the position of the nose and eyes.

 Making Decisions: CNN determines whose face it is now watching by comparing the
identified face to a database of known faces after features are identified. It is comparable
to aligning a jigsaw piece correctly.

4. Implementation Needs and Conditions

 Hardware’s and mechanic: CNN  Data and analytics: A substantial
training and inference require strong collection of tagged face photographs is
computing resources, such as required in order to train a trustworthy
specialised hardware accelerators, facial recognition system.
among other components.
 Expertise: Knowledge of CNN
 Software and developer: Deep architectural design, deep learning
learning frameworks like Tensor Flow concepts, and data preprocessing
provide libraries and tools that enable techniques are necessary for developing
the development and training of CNN and optimising the model.
models.

4. Conclusion
Because convolutional neural networks (CNNs) can automatically extract and recognise
information from visual input, they are incredibly useful tools in machine learning, particularly
in facial recognition systems. CNNs offer precise and practical facial recognition solutions for
many real-world situations by using vast datasets and advanced computing powers.