astraIn talk about large-scale biometric systems, we hear a lot about algorithm performance: speed, accuracy, template size, false accept and reject rates, ROC and DET curves, etc. These are critical to knowing if a biometric algorithm is up to the task at hand.

But we hear less about how these systems achieve their massive scale; their ability to compare a “probe” to a biometric “gallery” containing tens of millions of biometric records and generate an accurate result within seconds. Accurately comparing just two biometric samples reliably presents its own challenges; doing so millions of times per second is a very different feat altogether.

Modern, large-scale biometric search technology has more in common with more lauded achievements of the information age than meets the eye. After all, biometrics have been about “big data” and “the cloud” for decades. Taking the FBI as a case in point, its computers were used to search fingerprint files as early as 1980, and it provided fingerprint search as a service to local law enforcement back in 1999.

Achieving acceptable response times and search frequency are important motivators for scalability, especially as biometric search is increasingly used beyond law enforcement for civil and commercial applications. When the subject of a search is sitting in a jail cell, getting results in a few hours might be acceptable. But when they’re waiting in a long customs line after an overnight flight, it’s a different story altogether; response time is measured in seconds, not hours.

A challenge that drives the need for scalability is that biometric galleries are always growing, often at a rapid pace. If you’ve ever tried to manage your personal photo collection as it grows and grows, you might have some appreciation of the prospect of managing millions of biometric identity records. A biometric gallery can grow from thousands to millions of samples in a few years’ time. Plus, the larger a biometric database gets, the more data that is needed per enrollment to maintain the same accuracy. As a fingerprint search gallery grows into the millions, it might go from requiring two fingerprints per enrollment to ten. It’s prudent to carefully plan for that kind of growth.

Modern biometric systems should be not only scalable, but also sufficiently flexible to accommodate future innovations and requirements. A system owner will inevitably want to enhance their platform by adding new biometric modalities, replacing older algorithms, or even by adding “intramodal” algorithms from different vendors alongside the ones they already have. Achieving this degree of flexibility with a large-scale system is a tall order; generally, when you want to switch to algorithms from a different vendor, you have to start from scratch, and when you want to use different algorithms simultaneously in the same system, you need discrete, stovepiped matching systems.

Aware’s Astra™ is the newest, most modern biometric matching platform available on the market. It brings new features and capabilities to large-scale biometrics, in part by leveraging concepts and open source software built for other recent cloud-based, big data-centric challenges. Astra is essentially a biometrics-optimized cluster computing platform that distributes biometric algorithms and data across multiple computing nodes.

One of several innovative features of Astra is that it’s designed to be independent of the matching algorithms. Astra can utilize Aware’s Nexa™ fingerprint, face, and iris algorithms, but can also support algorithms from different suppliers, such as for voice biometrics. This gives customers flexibility in algorithm selection and also keeps the door open to unanticipated capabilities down the road.

Another innovative feature of Astra is that it utilizes several open source cluster computing and data management software components that are well supported and broadly field-proven in a variety of very large-scale platforms run by household-name technology companies. Out of the box, it’s able to handle issues encountered in other applications that we may not always foresee in the biometrics arena. It took several years of work to optimize the platform for scalable biometric search.

Scalability isn’t always the biggest motivator, but fault tolerance, robust failover, and disaster recovery are rarely negotiable requirements. With cluster computing, scalability is the cake and redundancy is the icing. Support for seamless failover is built into Astra. When problems happen on one machine, functions automatically move to another, like on-demand all-wheel drive. Underlying computing nodes can even be geographically distributed.

Yet another unique feature of Astra is its ability to run fuzzy text-based identity searches and analytics. Aware’s Inquire™ algorithms perform probabilistic comparisons and linkages between text-based identity records, which can achieve some powerful capabilities such as advanced text-based pre-filtering of a biometric search, data characterization and quality analysis, identity resolution, and relationship detection. These are all highly complementary to biometric search, and can contribute materially to the end-goal of identification.

Astra is ready-made to run in a private or public cloud. Running software applications in a cloud essentially means abstracting it from underlying hardware. Astra is designed to run independently of hardware and on as few or as many machines as needed to achieve the desired performance. Other applications can be easily integrated with Astra through a service-oriented web interface. A dashboard, system monitoring features, and alert services are all provided to give visibility into the production cluster and make it easier to manage and optimize.

Astra is truly a next-generation biometric search solution built for the cloud from the ground up, optimizing powerful, mass market-scale open source platforms for the unique challenges of highly scalable biometric search and authentication. In doing so, it delivers important features that set a new benchmark for large-scale biometrics.