The Silent Killer of Facial Recognition Software

facial-recognition-software.jpg

It seems as if facial recognition software is all the rage today. New deployments are rolling out across the globe and additional vendors and apps continue to hit the market. Rumors are even swirling that the next iPhone will embrace face recognition software for user authentication, replacing the TouchID we’ve all grown to love.

Afterall, what’s not to like about facial recognition software? It’s easy to use, non-intrusive, it’s the natural way humans identify each other, cameras are ubiquitous, and our faces are constantly on display.

So, what could possibly go wrong when you decide to implement your own facial recognition software solution?

Even though there are a number of considerations that need to be addressed, there remains one area that could potentially undermine your program’s success - the environment in which it is deployed.

Sure, all biometric modalities have their faults and require extra attention when deploying a solution, but facial recognition software is more apt to encounter unique issues based on some of the very benefits mentioned above.

The environment is the most critical element when designing and deploying facial recognition software. The variance and complexities presented require you to look beyond simple accuracy numbers touted by a given vendor. Those results are too often derived from tests run with a small population in a controlled environment. What happens when you throw the technology into your unique and challenging setting?

Even though facial recognition software has grown by leaps and bounds over the last 20+ years, you have to understand the technical and operational impacts the environment may present.

Here are several interrelating environmental factors you should be aware of when thinking about deploying your next facial recognition system.

Facility

Where are you deploying your camera systems? Will it be in a pristine environment such as a new airport or will it be in the dingy recesses of a prison?

While it may seem trivial that a camera will work in any environment, there are a number of inherent challenges you’ll encounter based solely on the type of building where the system is deployed. Is lighting generally poor? Does the target population typically interface with technology? Is it excessively dirty and dusty?

If you’ve seen one facility, you’ve seen one facility. Each will provide its own set of challenges.

Indoor vs. outdoor

Indoor facial biometric systems are typically much easier to implement. But, what if you want to deploy your biometric system outside? There are a number of considerations you’ll have to take into account.

Are the devices ruggedized and able to withstand the demands of the weather? If not, you may need to deploy the device inside an enclosure meant to protect it from the elements. When you do that you may negatively impact the results of the system. Glare from protective glass or reflections on the face may throw off face detection software. The heat inside the box may warrant additional heating and cooling controls.

The users of the outdoor facial system may also inadvertently impact the accuracy of the system. They may be wearing hats and/or sunglasses. They may be squinting and looking away given the location of the sun.

Will you need to deploy a canopy or awning to cover the devices? Even when doing this the sheer magnitude of background ambient lighting may impact the quality of photos captured. This, in turn, may impact biometric matching depending on the facial recognition software implemented. The location of the sun and how direct light hits the camera may also cause unnecessary capture failures and system errors.

Lighting

Perhaps one of the biggest, and most consistent, environmental issues with facial recognition software is lighting. While vendors continue to refine face capture capabilities and the technology has improved tremendously, lighting remains a challenge.

Do you have fluorescent lighting overhead that casts excessive shadows on the face? Is the light source directly behind the user? Are there large windows that provide variable lighting throughout the day? What are the lighting differences between night and day and how does that impact your system?

If you can master the consistency and placement of lighting, your system will have a much higher probability of success.

Image inconsistency

How consistent are the live images you capture to the images used as reference for matching? Is your probe image taken at an angle in poor lighting and then used to match against a pristine photo that was taken under ideal conditions? These variances will impact facial matching accuracy.

Also, be cognizant when reviewing facial biometric software vendors. Many times they will show the enrollment and subsequent match of an individual on the same day in the same environment. While this shows core functionality of the system it unfortunately covers up potential issues when deployed in the real world. The image captured in the field will vary, sometimes greatly, from the image captured during the initial face enrollment. This includes age, presentation, location, and orientation.

Physical location of cameras

How are you planning on deploying your cameras? How much room do you have? How will the user’s face be presented to the camera? Will it be an active process with direct interaction? Or will it be completely passive?

What will the facial capture area look like? Will the camera be fixed focus at a certain depth? What are the variances of height needed to capture all of the users targeted by the system?

How you physically deploy cameras and how users interact with them will directly impact the success of your facial recognition software.

Background

Will the background of your images be controlled or will it be complex and dynamic? Will one person be authenticated at a time or will other users be present.

Given the inherently larger capture area of a facial biometric system, you might pick up additional items in the camera’s field of view. This could include additional faces, users who aren’t in your target population, or even objects that trick the facial recognition software into capturing what it believes is a face.

In summary

These are just a few of the environmental questions you need to address. While facial recognition software shows great promise and continues to get better every day, there are still inherent risks associated with the technology. It is not perfect and you can expect a level of degradation based on where and how you use it.

When deciding on what you want to buy and ultimately deploy, make sure you look beyond simple match rates. Look deeper at your operational environment and identify what may silently kill your implementation.