Face Recognition | Facial Recognition using Deep Learning
Introduction:
Research in the domain of Facial Recognition or Face Recognition Systems has been conducted now for almost 50 years.
Over the years, it has found numerous practical applications in the area of biometrics, law enforcement, surveillance, access control, smart cards, and information security.
Facial recognition is a software that maps an individual’s facial features mathematically and stores the data as a faceprint. Nowadays, computer vision with deep learning is extensively used for making a face recognition system.
Existing Access Control methods and issues:
Presently fingerprint recognition, RFID ID cards, password-based security systems are the majorly used techniques for access control.
These legacy systems have a fundamental requirement of being hardware driven and hardware dependent. Another issue that these conventional systems face, is in cases where integrated access control has to be enabled at multiple interfaces.
e.g. an access control system which requires a two-step authentication, the first at the entrance door and the second at a computer system.
In such cases, facial recognition systems can help the user do away with unwanted additional hardware and provide better system integration by providing consistent access control throughout the organization.
Why use facial recognition over legacy systems
- Removes physical interaction with hardware: It requires no physical interaction on behalf of the user. The damage to fingerprint recognition sensors, for example, have a scheduled maintenance need due to wear and tear. A camera, on the other hand, reduces this added expenditure.
- Secure and Reliable: Facial recognition technique for access control is highly secure and reliable. It is very difficult for someone to breach the system since the only hardware used is a camera. Interpolated 3D face recognition has been seen to give even better results.
- Minimal dependence on hardware: The system can be installed very quickly and can be easily integrated with existing hardware infrastructure, existing cameras.
- Minimal maintenance cost: Since the hardware used for physical interface is the camera, the overall system becomes very easy and economical to maintain.
- Easy integration across multiple locations and entities: The basic camera enabled system can easily be implemented across diverse locations using a cloud-based approach. A stand-alone module of this facial recognition can be used for both granting access at physical barriers (labs, sensitive locations) and sensitive computer systems.
The science behind Facial Recognition
Every face has numerous, distinguishable landmarks, the different peaks, and valleys that make up facial features.
These are called nodal points. Each human face has approximately 80 nodal points.
Some critical data points are :
Distance between the eyes
Width of the nose
Depth of the eye sockets
The shape of the cheekbones
The length of the jaw line
These nodal points are measured creating a numerical code, called a faceprint, representing the face in the database.
Facial recognition system internally goes through various steps.
Step 1: Detection of faces
Since deep learning is a more accurate method of face detection as compared to HOG + Linear SVM object detector hence it is used to locate the presence and location of a face on a frame.
The face is not identified at this stage. The algorithm then extracts the 128-d feature vectors (called “embeddings”) that quantify each face in an image. Tracking can be done in the video once the face is detected.
Step 2: Feature Extraction:
A convolutional neural network is majorly used to extract features from input images. The neural network computes the 128-d embeddings for each face and then improves the weights of the network (via the triplet loss function).
Triplet loss works on minimizing the distance from positive examples while maximizing the distance from negative examples. In embedding space, Euclidean distance directly corresponds to face similarity.
In this manner, the network is able to learn to quantify faces and return highly discriminating embeddings suitable for face recognition.
Step 3: Train Facial Recognition model:
A machine learning model is trained by providing an ML algorithm (that is, the learning algorithm) such as an SVM, k-NN classifier, Random Forest, etc. with training data to learn from. These ML algorithms help the system in identifying the faces of people.
Step 4: Recognition
Recognition of face is done by extracting the face from the query image and generating a 128-D vector of the face. Then, we use that vector to decide if the features extracted from the new sample are matching or not.
Alignment:
Face recognition system results depends upon how the face is aligned towards the camera. Hence recognition accuracy can be increased by aligning the face based on translation, rotation, and scale.
Dlib library can be used to detect facial landmarks. These facial landmarks further used to align face properly using openCV.
New Challenges met by Facial Recognition (Applications)
General Identity Verification:
Facial recognition can play important role in identity verification like Electoral Registration, banking, electronic commerce, national IDs, passports, employee IDs.
Hardware Security:
With the increase in dependency of cellphones, computers for confidential work it is very important to secure the hardware. 3D facial recognition biometric can be used to elevate the level of phone and computer security.
Criminal Identification
Face recognition can help quickly identify suspects in the field. Face recognition system can be installed in vulnerable areas to quickly identify persons of interest.
Since it only requires camera hence identification can be done from anywhere just using a cell phone camera and a centralized system.
Airport Security
Places like international airports are more likely to have the movement of people of interest trying to evade enforcement authorities. This presents a security challenge as well as an opportunity to catch criminals by scanning their face.
Offices
In these modern times, office security is the need of the hour for the development of the market as a whole which in turn will help in the development of our cities industries.
So the concept of facial recognition to gain access to the office is an idea which is used to make our working area more secure. Face recognition can be used in office door access control, hardware access control and many more.
Scope for future work
Face recognition systems used today work under constrained conditions, these systems work much better with frontal mug-shot images with constant lighting.
All current face recognition algorithms fail under the varying environmental conditions. Next generation face recognition systems will need to recognize people’s face in real-time and in much less constrained situations.
Identification systems must be robust in natural environments, in the presence of noise and illumination changes, cannot rely on a single modality, so the fusion with other modalities is the way this area will progress.