By: Jesse Davis West February 07, 2019

Here at FaceFirst, we have the pleasure of working with a diverse set of technology partners to create powerful solutions powered by facial recognition. As face recognition stands poised to revolutionize customer loyalty and public safety, companies reach out to us regularly asking about about our facial recognition platform. We thought it would be beneficial to provide a list of the most important questions that companies should ask when choosing facial recognition partners to work with. The answers to the following ten questions should provide ample insight into whether it is worth pursuing a partnership with a face recognition company.

Scalability

Do you offer fast and flexible deployment options?

One of the most important keys to scalability is ease of deployment. When selecting a facial recognition vendor, be sure to ask what deployment options they offer. As an example, can they handle cloud deployments? Can they work with edge servers? It’s important to ensure that they offer deployment options that meet your needs. It’s also important to ask about deployment speed. The best facial recognition vendors should offer fast deployment options.

Can the platform handle a large database?

This question is especially important for companies or partners that wish to deploy facial recognition for enterprise use cases. While some facial recognition systems are designed to work with smaller databases, FaceFirst has made it possible to instantly match against databases of virtually any size.

How does the platform handle surveillance across multiple locations?

One of the most powerful benefits of face recognition is creating a network effect across multiple locations. This means that, for example, a retail chain could share a single database across a variety of stores. If a dishonest customer steals from one store, is enrolled in a face recognition database and then enters a different store location, security professionals at the second location would still be able to instantly identify the shoplifter. This is a key area where not all face recognition platforms are equal. With some platforms, there can be immense variations in terms of speed, accuracy, scalability and capabilities across locations.

Integration

Do you offer an API/SDK?

The best facial recognition systems have well-documented APIs and SDKs that allow for integration with various third-party technologies. We’ve found that customers often like to use facial recognition for custom use cases, and having APIs and software development kits can help customers cobble together their ideal facial recognition solution.

How easily do you integrate with video management systems?

An important facial recognition capability is the ability to plug into an existing video management system to see face recognition matches, analytics and locations and even enroll directly from the VMS. Creating deep and impactful VMS integrations is essential for many security situation, so having a true platform integration framework that makes it extremely easy to build apps or similar integration layers should be high on the list.

Can the system integrate with mobile devices?

We all know that the iPhone offers uses face recognition to protect personal data.  But there are all kinds of mobile use cases of facial recognition ranging from police identifying suspects in the field to instantly recognizing travelers at transportation centers. Depending on the use case, it may be essential to have facial recognition capabilities on mobile devices. Be sure to ask if the vendor can integrate with mobile devices either by offering an application or APIs that enable mobile integration.

Communications & Alerting

How are alerts or other communications handled?

Whether facial recognition is being used for security purposes or to enhance customer experiences, real-time alerting is essential in order to optimize customer experiences and deter threats. It’s therefore essential to ask if a facial recognition provider offers real-time alerting and to test it in the wild if possible. In addition, flexibility — the ability to quickly set up and configure individual and group alerts through rich text messages, email, in apps and through other data collections systems, is important.

Privacy

What steps have been taken to protect privacy?

Facial recognition platforms should be designed from the ground up to protect privacy. While we can’t speak for other platforms, FaceFirst practices a privacy by design methodology that always ensures that personally identifiable information (PII) and biometric information is collected and stored, securely and with encryption. All the personal information stored in the FaceFirst system is protected, secured and isolated in multiple ways. We fully support and abide by the data privacy principles established by applicable local privacy laws and regulations. As part of our commitment to data protection, we conduct external security audits and independent security testing on a regular basis.

Accuracy

Is the solution accurate in the wild?

Facial recognition companies often tune their algorithms to work well in controlled lab environments or to succeed at industry benchmark tests such as NIST. However, it is far more important that the face recognition algorithm works well in the wild. This is especially true for surveillance applications that need to match at distances, in challenging lighting conditions or with occlusions.

We can’t speak for other platforms but FaceFirst  delivers exceptional accuracy at distances of up to 100 feet, and increasingly works well with less-than-ideal facial angles (like quarter- or half-profile images) against large databases. FaceFirst is also proven and tested with a variety of occlusions, including:

  • Hats
  • Sunglasses
  • Beards
  • Scarves
  • Partial profiles

How much has accuracy improved in recent years?

Facial recognition companies should be constantly tuning their algorithm to improve accuracy. It’s therefore beneficial for companies to ask prospective face recognition partners how accuracy has improved over the past two years. As an example, FaceFirst has been consistently retooling our algorithm over the past two years and we’ve seen dramatic improvements to accuracy (as well as speed) as a result.

Is the facial recognition algorithm trained on a diverse data set?

It is essential for facial recognition algorithms to be trained on diverse data sets. This is because using a diverse set of ground truth images is the best way to prevent bias in terms of gender, age, race, nationality and other factors. While we can’t speak for all facial recognition platforms, we take care to train our algorithm with a diverse set of ground truth images. As a result, we’ve achieved 99.97% tested accuracy on highly diverse public data sets commonly used in face recognition.