You want to create a cool new app that uses facial recognition. The job should be easy. After all, crime shows on TV show facial recognition instantly and easily recognizing faces perfectly without error. Software Development Kits (SDKs) exist so dropping facial recognition into your app should take less than a day.

Well, developing a reliable app is not that easy.

Before building any app, it is important to understand the current state of facial recognition technology, the requirements and the limitations.

Current State of Facial Recognition

The original applications for facial recognition were designed to match faces to driver’s licenses and passport photos to other full frontal poses. Today, facial recognition has improved its accuracy dramatically for these types of comparisons. The technology has improved significantly for photos and poses that do not match those full frontal ideal lighting non smiling environments. However, the technology is still designed to best match current snapshots with good lighting and very little pose deviation in a current time frame.

Requirements of Facial Recognition SDKs

SDKs to match faces are heavy in terms of memory and processing. The algorithms analyze the pixels in the face and search for general shapes and features or landmarks.

Vendors have licensing agreements for facial recognition. There is open source code for face finding, but the actual face matching requires licensing be a reliable SDK vendor. Today, there exists only a handful of vendors who supply accurate SDKs. There exist some vendors who have poor reliability in their SDKs.

It is very easy to get started with the SDKs. The code is easy to install and the documentation is good. The weaknesses are that an application developer must understand the nuances, the uses and the limitations of the facial recognition technology. If you do not have expertise in the field of this biometric, your development time will be extended considerably. An SDK developer must still code around the limitations of the technology instead of the limitation of the SDK.

Limitations of Facial Recognition

If not used properly, your app can produce false positives (the SDK claims that the person in the photo is person A but in reality it is not) and false negatives (the SDK claims that the person in the photo is not person A but in reality it is).

What can cause these errors?

• Poor enrollment. The quality of the first face enroll dramatically impacts the accuracy. How do you insure a good enrollment?
• Pose quality photos
• Pose deviations
• Occlusion
• Aging
• Clustering (geometrically, many people look alike)

Reducing your Learning Curve

If you wish to develop an app or service using facial recognition, consider using Initial Point to build the code and/or framework for you. Since Initial Point has developed applications such as PicsMatch, we have experience in developing apps and services using facial recognition. We can provide development services as well as tips and techniques to make your application much more successful.

Before beginning to develop your app, you should consider the following issues:

• Is the application one to one or one to many?
• How big is your population that you will examine?
• Do you need clustering?
• What techniques will you use to obtain the BEST enrollments?
• What are you prepared to pay in licensing fees?
• How will you handle accuracy errors and photos with faces that do not meet SDK specifications?

If you need assistance in answering those questions or you would like to reduce your facial app development learning curve, then This e-mail address is being protected from spambots. You need JavaScript enabled to view it .