Facial scan and recognition is one of the leading biometric technologies that is preferred in many places.

Components of facial scan system:

The two main components of facial scan technology are the face location engine and the face recognition engine. The face location engine tracks faces in a field and the face recognition engine compares the tracked faces. The main intent of the technology is to enroll, verify and identify the facial images – which are to be acquired through different methods such as static photographs, web cameras and surveillance cameras.

Due to its complex functionality, the facial scan technology is most likely to be used in large-scale identification sectors or surveillance purposes.

How this works:

There are five standard steps involved in facial scan technology. They are,

  • Image acquisition – High resolution cameras are required to acquire images. Distance from the camera reduces the size and resolution of the image and the user not looking directly into the camera affects the image acquisition efficiency of certain technologies. Ideally, images having more than 100 pixels of height and width each helps the technology perform well.
  • Image processing – Normally, the acquired images are cropped into ovoid shapes and are turned into grayscale. Characteristics such as the middle of the eyes are located and used as a reference to straighten or position the image along an axis. Most facial scan systems acquire from 3 to 100 images to enroll an individual.
  • Distinctive characteristic location – Characteristics that are least likely to change over time (such as upper ridges of eye sockets, areas around cheekbones, nose shape, etc.) are the most utilized. Yet the facial scan systems have not developed enough to identify a person even when ttheir expressions are changed.
  • Template creation – The enrollment templates are created from many processed images. Its size may vary from less than 100 bytes to 3K bytes. These templates are merely representations of data located during feature location.
  • Matching – Confidence levels are assigned to the strength of each match attempt between a match template and an enrollment template. If the score surpasses a certain level, it is deemed as a match. But since this is not as effective as finger or iris scan, identifying a specific person from a large database is hectic as the system returns all the potential matches. The actual match would then be determined by a human opreator if possible.

Competing technologies/algorithms:

Four of the different facial scan methods employed by deployers to verify people include Eigenface, feature analysis, neural network and automatic face processing.


Eigenface is a technology patented at MIT which translates to “one’s own face.” This makes use of two-dimensional, grayscale facial images to create templates during enrollment and verification. Distinctive facial characteristics can be reconstructed by locating those features from around 100 – 125 Eigenfaces. Upon enrollment, a subject’s facial image is represented using a combination of various Eigenfaces and mapped to a series of coefficients. A user’s live template is compared against the enrolled template to determine the coefficient variation and the degree of variance, which will determine acceptance or rejection.

Feature Analysis:

Feature analysis is most widely used and this technology is similar to Eigenface but can accommodate changes in appearance or aspects such as smiling/frowning. Visionics, a facial recognition company, uses Local Feature Analysis (LFA), which is described as a reduction of facial features to an “irreducible set of building elements.” Dozens of templates from different regions of the face are derived and extracted, including their location. It is implied that a light movement of a feature will be accompanied by similar movement of adjacent features. Feature analysis can accommodate up to 25 degrees in the horizontal plane and 15 degrees in the vertical plane. A facial image from a distance of 3 feet is considered ideal for enrollment.

Neural Network:

Algorithms are employed to compare live and enrolled faces and neural systems are designed to learn which features are most effective within the pool of users. Features from the enrollment and verification faces vote on whether there is a match and an incorrect vote prompts the algorithm to modify the weight given to certain facial features. So, the neural network decides and adjusts the features used. This technology learns and gets better over time, which may be able to reduce the time-based performance problems found in these kinds of systems. But this system is not ideal for surveillance applications as it does not provide a well-suited environment for neural net enrollment.

Automatic Face Processing:

Automatic Face Processing primarily uses distances and distance ratios. These ratios are acquired from between the eyes, end of nose, corners of mouth, etc. this technology is much more effective in dimly lit, frontal image-capture situations. This is used in booking station applications.

Facial scan strengths:

  • Ability to operate without physical contact.
  • Ability to enroll static images.
  • No need for specialised equipment or technology to function.

Facial scan weaknesses:

  • Changes in physiological characteristics reduce accuracy.
  • Privacy breach may occur.
  • Non-ideal acquisition environment leads to failure of enrollment and verification.

Thus the facial scan technology and its different algorithms were discussed in detail.

Click to read about coffee

Click to read about accuracy in biometrics