The Intricate Workings of Facial Recognition Systems

A common term that you are likely to come across a lot in the world of security is facial recognition. This technology continues to grow and baffle everyone who interacts with these systems. Facial recognition is a product of applied machine learning that can sense and identify human faces. Identifying human faces has been a major challenge for computer scientists for eons. However, through ground-breaking research, advancements have been made and great ones at that.

In the beginning…

The buzz around facial recognition software might lead you to believe that it is new technology, but this tech has been around for a while. Algorithmic work in detecting faces can be traced back to somewhere around the turn of the millennium, when the Viola-Jones Object Detection Framework was published. Though it was initially used to identify objects within images, it gradually shifted focus to facial recognition. One of the reasons why the algorithm was popular was because it was fast. However, the training process was slow.

Real progress and results started to show in the 2010s. This was after the introduction of Convolutional Neural Networks as an approach to perform facial detection. The capacity to provide raw processing power and gigantic system memories made it easy for cloud computing by Infrastructure as a Service (IaaS) providers. This provided capacity for computers to consistently beat humans when it comes to facial recognition, even when large numbers of faces were involved. So how exactly does it work?

·         Detecting and tracking

This pre-processing stage is responsible for detecting and tracking faces in a given image or video file. During tracking, certain parts, features, and expressions on a face are needed. Sometimes a whole process of facial expression recognition software is needed to complete this process.  

·         Alignment

Facial recognition is the compounded nature of faces in a given image or video that do not follow any rules. A person may have been zoomed in or out while peeking from a wall or a tree, or they may give a side profile making the problem of face detection difficult. Face alignment comes in as a corrective measure to identify the video's lines and the parameters for facial features. 

·         Feature extraction

During this stage, the individual features of the face such as eyes, nose, chin, lips, etc. are extracted to be put to use by algorithms in the subsequent stages.

·         Feature matching  

Once all of your face's essential features have been collected, the process of matching now begins. This stage is also known as a classification because the algorithm can now individually identify a face. The extraction matched against the given database helps work out the identity of the person.