Face recognition is a popular topic in the world of computer vision. It’s also a topic with many misconceptions and myths surrounding it. This post will tackle the basics of online face recognition and demystify some common myths about what face recognition can do and how accurate it truly is. We will discuss why accuracy affects a business’s decisions regarding what types of images a business should collect from customers if a business wants to use this technology for marketing or other uses.
How does online face recognition work?
For a more detailed explanation of how online face recognition works, the Industry has a guide that can help. This section explains the steps in the process and gives a business some basic information on how it all comes together.
It’s essential to remember facial recognition technology is still relatively new and evolving—so many of these details are not set in stone yet. But here are some basics:
- Facial images are acquired through cameras and then processed by specific algorithms (the software). The result is an output file with unique identifiers for each person in an image library. This data can be used later when comparing faces against each other or stored records from previous visits or events.
Face detection vs face recognition: what’s the difference?
Face detection and face recognition are separate processes with vastly different uses. Face detection is a passive process, whereas face recognition is active.
Face detection can automatically detect and track faces in video streams. It can also be used with other processes—such as object tracking or background subtraction—to track only the objects of interest while ignoring everything else (see image below). Face recognition, on the other hand, is designed primarily for identifying people within photos or videos based on their appearance (and not just their position within the said image).
What are the uses of online face recognition?
Online face recognition can be used to:
- Identify a person in a photo.
- Identify a face in a video.
- Identify faces in video streams, such as the ones created by surveillance cameras.
- Identify faces in images and videos (including those taken with mobile phones).
What are some standard techniques & algorithms for online face recognition?
A face detector is a machine learning algorithm that detects whether or not an image contains a face. The most common technique is Haar-Feature, which uses Haar-like features. It has been demonstrated to be very effective in detecting frontal faces but can also be applied to profile and 3/4 views of faces.
Once all the faces within an image are seen, they will usually not line up perfectly because they are captured from different perspectives at different distances, etc. Thus one task is to register each face so that they all appear in the same position within the image plane. This process relies heavily on training datasets where multiple examples of each subject have been aligned manually by human annotators before being fed into a business model for training purposes (e.g., using TensorFlow).
A business’s goal here would be to develop a classifier algorithm. Such as a Support Vector Machine (SVM) that predicts whether or not someone should be categorized as having either male or female gender based on their facial attributes like nose size/shape, cheekbones height/width ratio, etc.
How accurate is online face recognition?
Online face recognition is not 100% accurate. The algorithms companies use to identify people are pretty good, but they’re not perfect.
- To recognize a face, the first step in the recognition process is identifying the face. It can be done by analyzing local features or critical points or detecting a particular shape (i.e., oval).
- Face recognition: Once a business has identified that there is indeed a face present in the picture, a business must then recognize it as belonging to one individual rather than another. It requires two steps: performing feature extraction on each detected face and matching extracted features against reference data stored in a database containing information on many people’s faces.
- Face alignment/verification/identification: Once a business has determined which person is present in an image, this also allows for verification of their identity against other pictures taken at different times or angles (alignment).
- If a business knows whose face it is but needs confirmation that they are who they say they are when accessing a business site online (verification), then using facial biometric techniques such as iris scan or fingerprinting can be used instead of passwords. Since those methods are much harder to hack than passwords, fewer fraudsters try their luck at getting into our systems without permission.
The guide has been an excellent resource for learning about the techniques and algorithms used in online face recognition. For a deeper understanding of this technology, readers should read it to understand better what it does and what it can be used for.