Blog Archives | ROC https://roc.ai/category/blog/ Rank One develops industry-leading, American-made computer vision solutions that leverage Artifical Intelligence and make the world safer and more convenient. Mon, 12 Jun 2023 17:15:32 +0000 en-US hourly 1 https://wordpress.org/?v=6.7.1 https://roc.ai/wp-content/uploads/2024/02/cropped-Group-44-1-32x32.png Blog Archives | ROC https://roc.ai/category/blog/ 32 32 Unpacking AI: How Deep Learning Algorithms are Made https://roc.ai/2023/06/12/unpacking-ai-how-deep-learning-algorithms-are-made/ Mon, 12 Jun 2023 17:15:32 +0000 https://roc.ai/?p=10013 The post Unpacking AI: How Deep Learning Algorithms are Made appeared first on ROC.

]]>

The hottest buzzword of 2023, ‘artificial intelligence,’ persists as an extremely broad term with near-infinite interpretations and possibilities. Despite the attention, most AI algorithm development processes remain shrouded in mystery.

At Rank One Computing (ROC.ai), we build smarter, faster solutions for the world’s most pressing challenges, with greater levels of transparency. As machine learning continues to dominate news cycles, we share a special look behind the scenes at how deep learning algorithms are created and integrated into practical applications.

At their core, artificial intelligence (AI) and machine learning (ML) algorithms train on vast amounts of data to inform better decisions, predict likely outcomes, and even create new data or content. 

Analytic AI algorithms – like those made here at ROC.ai – are trained to compare new data against statistical models trained on large volumes of existing data in order to authenticate, examine, match, predict, identify outliers, or measure traits.

Generative AI algorithms – like ChatGPT, Grammarly, and Midjourney – are trained to produce new multimedia content within the parameters of models trained on existing media, in order to execute user-provided scenarios or ‘prompts.’

In all types of AI, the source and treatment of these large training datasets are critical to the quality and performance of the resulting AI algorithms. While generative AI tools continue to raise questions, concerns, and even bans across the country, ROC.ai’s computer vision and biometric analytic algorithms continue to raise global standards for accuracy, efficiency, and ethical use across modalities – including face recognition, fingerprint matching, object detection, license plate recognition, and more.

So how are these game-changing algorithms developed and refined over time? In this article, gain insight and transparency on the four key steps – data development, algorithm development, algorithm integration, and customer support.

Step 1: Data Development – The Foundation of AI Algorithms

The process begins with data development, a crucial step in creating next-generation deep learning models. These models require vast amounts of labeled training data to learn model parameters that generalize across a range of operational use-cases and conditions. Though plenty of source data exists, ethical data sourcing is essential to ensure unbiased algorithms. Human data validation of collected data plays a critical role in minimizing algorithmic biases and improving overall accuracy. By carefully curating and validating the training data, ROC.ai developers establish a solid foundation for each algorithm.

Step 2: Algorithm Development – Unleashing the Power of Patterns

Training high-quality algorithms requires deep knowledge of statistical pattern recognition principles (as captured in the seminal textbook Pattern Classification), specialized hardware infrastructure, and powerful Graphics Processing Units (GPUs). The training of highly accurate algorithms can be a time-consuming endeavor, often spanning days, weeks, or even months. After a candidate algorithm is trained, test and evaluation datasets measure both absolute accuracy in real-world use cases, and relative accuracy through comparing the performance of different algorithms (e.g., prior releases to ensure an improvement is being delivered.)

Step 3: Algorithm Integration – Bridging the Gap Between Research and Deployment

Once the algorithms have been developed, the next step is software integration. This phase entails porting the trained computer vision models into deployable software libraries (like ROC SDK) and systems that can operate across various hardware architectures and software operating systems. These software libraries require careful creation of APIs (Application Programming Interfaces) to ensure software developers using the models embedded in these libraries can easily build fault tolerant systems.

Once the integration process is complete, we conduct extensive testing to validate the integration process and ensure that the computer vision models perform consistently in their integrated form with how they performed in the algorithm development environment. 

Step 4: Customer Support – Enabling Seamless Communication and Feedback

Integrators and customers of deployed algorithms play a pivotal role in the final step of the process: customer support. We build and maintain direct communication channels with integrators to provide feedback on the algorithm’s performance “in the field.” Such feedback can often include data exhibiting a specific set of conditions. This feedback and data can significantly impact future research cycles and aid in continuous improvement. 

Looking Beyond the Buzz

As the buzz around AI continues to build, it’s critical to understand that not all AI is created equal – meticulous development cycles and data collection methods make all the difference.  From data cultivation through testing, integration, and customer support, each step plays a vital role in bringing these cutting-edge technologies to life. As the demand for intelligent computer vision solutions continues to rise, single-source vendors like ROC.ai will lead the charge, driving innovation and revolutionizing industries with expertise in this rapidly evolving field.

Ready to ROC? Let’s Talk.

Rank One Computing builds faster, more accurate, reliable tools for a safer, smarter world. Our American-made biometric and computer vision algorithms are trusted by public safety and security leaders nationwide to protect critical public and private infrastructure. Reach out to learn more about how our industry-leading technology can enhance security operations for your organization.

The post Unpacking AI: How Deep Learning Algorithms are Made appeared first on ROC.

]]>
Looking Back at Boston Marathon Bombing: A Decade of Face Recognition Advancement https://roc.ai/2023/04/17/looking-back-at-boston-marathon-bombing-a-decade-of-face-recognition-advancement/ Mon, 17 Apr 2023 19:23:55 +0000 https://roc.ai/?p=9554 Ten years ago on April 15th, 2013, the city of […]

The post Looking Back at Boston Marathon Bombing: A Decade of Face Recognition Advancement appeared first on ROC.

]]>
Ten years ago on April 15th, 2013, the city of Boston was struck by an unforgettable act of terror. Two bombs exploded near the finish line of the Boston Marathon, killing three people and forever changing the lives of so many more. 

A decade later, Rank One Computing CEO Scott Swann and Co-founder and Chief Scientist Brendan Klare recall their unique perspectives, and reflect on the last ten years of technology and innovation inspired by the tragic incident.

Recalling The Incident

Swann, now an 18-year FBI veteran, supported the FBI Science & Technology Branch within the Director’s Office in Washington DC as the case unfolded. Klare, like many, initially followed the disturbing news from home, but was later recruited by Swann to help identify technology solutions to address significant challenges that emerged from the investigation.

Swann recalls the immediate aftermath of the event, “I was working with FBI executives on the 7th floor. It was like a command center – so chaotic – things were happening really fast,” said Swann, “A video came in with the first big tip of who these guys were. My colleagues loaded it up on an iPad to get it in front of the President within minutes.”

A tragic act of terror had been captured publicly by so many cameras from so many angles, but law enforcement agencies unfortunately lacked automated solutions to help identify and track down the suspects. For the first time at scale, the FBI took the unprecedented steps of opening their tip line for the public to submit photos and videos to aid the investigation.  The response was overwhelming, and the FBI was quickly inundated by terabytes of multimedia from private citizens who wanted to help.

“It was almost too much video data to manage effectively at that time. There were great FBI agents and law enforcement officials working the events and at the end this is what really led to breaking the case,” said Swann. “I think the overload of data just opened up a lot of ‘what if’ questions. We knew we had a gap after this incident that we needed to address.”

Conducting a Major Issue Study

In the wake of the Boston Marathon Bombing, Swann recognized a significant technology limitation and suggested the need for a broad overarching assessment and industry study. So the 18-year FBI veteran led the charge for a Major Issue Study to better understand the video analytic landscape across commercial and government sectors, and to develop a comprehensive roadmap for managing video more effectively in the future.

Familiar with Klare’s graduate research and early work in the lab of Dr. Anil Jain, Swann eagerly recruited him as a consultant on the project. The assessment explored the scope of video processing challenges across government and commercial organizations, ultimately identifying massive cross-industry technology gaps, and recommending a path for prioritization and investment moving forward. 

There were no existing solutions that brought together all the capabilities needed at scale. “Child exploitation, gang affiliation, terrorist networks, criminal activity – as you think about the collection, analysis, and dissemination of such information it was daunting” said Klare.

At the time, face recognition technology wasn’t built for unconstrained environments with unusual angles or lighting, and it certainly wasn’t built to handle terabytes of data in seconds.

“We founded Rank One Computing in 2015 to focus on both accuracy AND efficiency, after experiencing these processing limitations firsthand,” said Klare. A few years later, the tables turned when Klare recruited Swann to join Rank One Computing as CEO. “He was the clear choice. He is an immense professional with incredible focus and vision, who carries weight and drives action,” said Klare.

While Swann moved on from the FBI after completing the study, the FBI’s Multimedia Exploitation Unit and others went on to stand up an impressive set of capabilities well beyond those initial set of recommendations.   

“The post-event analysis of the Boston Marathon Bombing was my last big assignment at the FBI. The contributions Brendan and I made with the video study were effective in getting some attention to a gap in technology capabilities. Working in the FBI Director’s Office provided one of the best vantage points I could ever have in understanding the professionalism of the agents that handle these cases. Equipping these men and women with the best technology possible is a force multiplier in preventing future catastrophes and ensuring a rapid response to serve justice,” said Swann.

Fast Forward to Today

Since 2013, the field of biometric and computer vision technology has made incredible progress. Face recognition technology is now faster, more accurate, and more reliable than ever before, enabling even small local law enforcement agencies to identify potential suspects more efficiently.

ROC has remained positioned at the forefront of this progress, working to develop faster and more accurate algorithms that can be used to enhance safety and security in a variety of critical settings. As an organization, ROC works to apply the important lessons learned from real world scenarios to improve safety across the country and around the world, including in schools.

Currently in West Virginia public schools, ROC’s deployed system can automatically control building access, helping to keep unauthorized individuals out. “There’s nothing more critical than our next generation of leaders,” said Swann. “Face recognition is now the most accurate biometric technology available – overtaking both fingerprint and iris identification in recent years. Our solutions can now process near infinite volumes of data essentially instantly.”

As Klare explains, “We can now organize and track data sets with more than a million unique identities in no time at all, on any hardware, even in unconstrained environments. We can also automatically identify long guns. Our biometrics have been used by law enforcement agencies across the country to catch violent criminals.”

ROC SDK, a multimodal software development kit, powers the fastest, most accurate, scalable solutions for not only identifying criminal suspects using face recognition, but also tracking the movement of suspects through a crowd (“clustering”), and securing spaces with automated visitor identification and watchlist enforcement. 

And ROC Watch, a live video analytics platform, channels the power of ROC’s industry-leading algorithms into a convenient, intuitive SAAS solution, unlocking innovative safety tools for more community and commercial organizations.

But while we have come a long way in the past ten years, there is still much work to be done. As we reflect on the progress we have made, we must also acknowledge the challenges that lie ahead. Issues of privacy, bias, and misuse of technology continue to be a concern, and we share a responsibility with our customers to ensure that our solutions are developed and deployed in a responsible and ethical manner.

Released in 2021, ROC’s Code of Ethics was the first of its kind in the biometric and computer vision industries. They require customers to abide by the code as part of their licensing agreement.

“There’s a lot we couldn’t do back then that we can proudly do today, but there are also many things still not possible today when it comes to computer vision and biometric technologies. We have to find the balance between pushing against limitations and recognizing boundaries,” said Klare. “As we look back on the role today’s technology could have played in the aftermath of the Boston Marathon bombing, we are reminded of the incredible potential it holds to make our world a safer and more secure place.”

The post Looking Back at Boston Marathon Bombing: A Decade of Face Recognition Advancement appeared first on ROC.

]]>
The Pros and Cons of Face Recognition https://roc.ai/2023/03/29/the-pros-and-cons-of-face-recognition/ Wed, 29 Mar 2023 20:02:02 +0000 https://roc.ai/?p=9193 The post The Pros and Cons of Face Recognition appeared first on ROC.

]]>
The capabilities of face recognition algorithms have skyrocketed the last five years. With substantial gains comes substantial responsibility – and so now we must reexamine the role face recognition can play in the future of biometrics and identity technology, and closely consider potential limits that these technologies may never overcome.

Unlike fingerprint or iris recognition, facial appearances are public. In turn, facial images have become widely available – trillions of them scattered across the Internet, plus the endless hours of faces embedded in video streaming services. A nearly endless source of facial imagery exists in every corner of the digital world. Even relative to other non-biometric computer vision or machine learning classification tasks, the availability of facial appearance data vastly exceeds data available to train other classification algorithms. In many ways, due to the overwhelming legacy of widely distributed facial imagery, it’s likely that facial recognition will continue down this path to become the most accurate machine learning technology in the world.

In this article, we’ll explore classic considerations for facial biometric traits and discuss how they align with modern advances and use cases. Through this examination, we have identified the following “pros” and “cons” of face recognition technology:

 

Pros.

Cons.

Highly Accurate Challenges with Familial Similarity
Convenient Challenges with Cosmetic Modification
Default Method for Humans Risks of Spoofing

Accuracy

Though historically face recognition has been the least accurate of the “Big Three” biometrics (facial appearance, fingerprint ridge patterns, and iris texture), in recent years face recognition technology capabilities have surged to become the world’s most accurate biometric technology.

 

 

 

Comparing Biometric Accuracy (Face, Fingerprint, Iris)

Comparing biometric accuracy

Sources:
Face: NIST FRVT Ongoing, Oct 2022 – Visa Dataset
Iris: NIST IREX IX report, April 2018 – Table D1, Single Eye – NOTE: the current NIST IREX Ongoing does not include 1:1 measurements
Fingerprint: NIST PFT, Oct 2022 – MINEX III Dataset, Single Finger

 

While these error rates are not a perfect apples-to-apples comparison, error rates produced by face recognition algorithms are significantly lower than those measured by any single finger or iris sample. 

Fingerprint has long since been considered the gold-standard for biometric accuracy (outside of invasive methods like DNA), so many are often surprised to hear that face recognition error rates are often substantially lower.

A closer look at the last five years of accuracy testing reveals an exponential decrease in error rates.

Evolution of ROC.ai Face Recognition Accuracy, 2017 – 2023

Evolution of ROC.ai face recognition accuracy

Over the past 5 years, when operating at a False Match Rate of 1 in 1 Million, the False

Non-Match Rate (FNMR) error has decreased by over 50x

Source: https://pages.NIST.gov/frvt/reportcards/11/rankone_014.html

How did facial recognition become so accurate? 

 

Public biometric

The exponential accuracy progression of face recognition technology is largely due to the following two factors: 

  • Facial appearance has been the primary biometric trait throughout civilized human existence; and
  • Facial appearance is not private.

In terms of the privacy of facial imagery, historically speaking, our facial appearance is the single least private piece of information about ourselves. When meeting a stranger, people may exhibit reluctance to provide simple information such as their name. Yet, they will provide their facial appearance immediately to countless strangers in day-to-day interactions. When in a public setting, the right to privacy of facial appearance simply does not and cannot reasonably exist. In fact, in some of the most liberal countries on the planet (e.g,. France, Denmark) it is illegal for a person to conceal their face in public. 

A tangible example of our lack of facial privacy in public settings can be seen when attending professional and/or televised sporting events. People pay hefty sums of money to attend games, knowing that their facial appearance may be broadcast on TV to audiences of millions.

 

 

Facial Privacy in Public Settings Examples

sleeping man example

Fan Example 1 (Source)

surprised man example

Fan Example 2 (Source)

In Fan Example 1, a fan was sleeping at a nationally televised baseball game. TV broadcasters noticed the fan sleeping and aired it nationally while discussing him at length. The fan attempted to sue the broadcaster (ESPN), but ultimately couldn’t find valid legal grounds and his suit was dismissed. 

In Fan Example 2, a fan was stunned after his football team lost a game in the final seconds. The fan was briefly aired on TV, but it was enough for Internet memesters to turn the image into a meme that went viral. The fan never consented to his image being widely distributed across the Internet, but at the same time he had no recourse to prevent this distribution. 

In both cases, the fans never consented to be broadcast on national television, or to further be spread online. Nor does any other fan when shown in the background of such televised events. Regardless, the lack of inherent privacy to one’s facial appearance means there is no recourse to prevent sharing facial appearance – aside from not appearing in public. 

Though precedent does not protect privacy of facial appearance, privacy concerns still exist regarding facial appearance and facial recognition. In fact, emerging technologies like highly-accurate automated facial recognition can create significant privacy issues due to the highly public nature of facial appearance. 

 

Deep Learning and Convolutional Neural Networks 

In the last decade, a technological revolution rapidly advanced the world of computer vision and machine learning. Inspired by the human visual processing system, we call this technology, deep learning via convolutional neural networks. This technique applies highly-tuned kernel convolutions against matrices of image pixels to yield powerful feature representations. 

The number of parameters in these models usually falls in the order of hundreds of thousands to hundreds of millions. To learn the optimal model parameters requires yet another order of magnitude more imagery than the number of parameters, as dictated by the “rule of 10” phenomenon in machine learning.

When provided with sufficient data, knowledge about the classification problem domain, machine learning methods, and GPU/supercomputing hardware, algorithm models can be trained that deliver truly stunning accuracy. In some cases, such as face recognition, these models can significantly surpass human performance.

 

 

NIST Face Recognition Algorithm Matching Accuracy

NIST face recognition algorithm matching accuracy
The plot above is from the NIST FRVT vendor scorecard for ROC.ai. It shows the matching accuracy of the face recognition algorithm on the same pairs of images that expert facial examiners were given three months to study and determine whether or not they were a match. Expert examiners achieved roughly 95% accuracy on the task, while several automated algorithms can now achieve perfect 100% accuracy in a matter of seconds. Shown above, the automated algorithm achieves perfect accuracy, and a significant separation in facial similarity score between the genuine comparisons of the same persons (the first 12 plotted values with very high facial similarity scores) and the impostor comparisons of different persons (the next 8 plotted values with low facial similarity scores). 

Face recognition has benefited tremendously from the combination of expansive data and incredibly powerful deep learning toolkits, and we expect error rates should continue to decline for years to come. 

However, as discussed previously, face recognition will also struggle to cross some hard boundaries. Eventually the “Moore’s law”-like effect achieved by face recognition algorithms these past 5 years may taper out. 

 

Convenience 

In addition to accuracy, another chief benefit of face recognition technology is convenience. 

Using a face recognition system typically requires little effort. When paired with continuous authentication and real-time screening, they are often completely frictionless. When used as a method of access control, face recognition often requires less user effort and cooperation than fingerprint or iris recognition. 

Due to such convenience, face recognition is the only primary biometric trait that can be used successfully in a fully unconstrained manner. Indeed, accuracies on highly unconstrained benchmarks, like IARPA Janus, have gone from extremely low a decade ago, to now approaching the accuracy of other biometric traits operating in fully cooperative settings in a relatively short period of time. 

Examples of Unconstrained Face Images in the IARPA Janus Dataset

Examples of Unconstrained Face Images
Image source: B. Klare, B. Klein, E. Taborsky, A. Blanton, J. Cheney, K. Allen, P. Grother, A. Mah, and A.K. Jain,  “Pushing the Frontiers of Unconstrained Face Detection and Recognition: IARPA Janus Benchmark A.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015.
Convenience and ubiquity do come with a cost. While facial appearance is overwhelmingly public information, faces can also be linked to more sensitive private information. This consideration becomes especially important as face recognition accuracy and speed continue to advance.

 

Default Method for Humans

The most common method humans use to identify another person in day-to-day life is face recognition. Human face perception is such an important task – we have a large, dedicated region of the brain called the fusiform face area whose sole function is human face identification. 

It is natural for automated systems to prioritize compatibility with our manual, legacy methods. For this reason – inherent human understanding – face recognition is often preferable to other biometric traits.

Perhaps there is no better example of the human dependence of facial recognition than the Thatcher effect:

Example of the Thatcher effect

Example of the Thatcher effect 1
Example of the Thatcher effect 2
The two images above are the same, with the exception that the first image is flipped upside down. We can hardly perceive the manipulations in the inverted image due to the fact that inverted faces are not processed by the fusiform face area of the brain. The second image is immediately recognized as being altered and not an accurate face image.

One of the biggest reasons for society to continue investing in the use of properly developed automated face recognition technology is our inherent reliance on the facial biometric in our day-to-day lives. 

 

Limits of Face Recognition

While face recognition technology continues to deliver unprecedented convenience and accuracy, we must also address some fundamental limitations and challenges. 

Identical Twins

One of the more fundamental limits on face recognition is the challenge of identical twins and to a lesser extent familial relations. Though only accounting for roughly 0.3% of the population, identical twins in particular create a substantial challenge for automated face recognition algorithms. The reason is fairly obvious and illustrated by the following examples of pairs of identical twins:

two children example
Source: https://flic.kr/p/2naHVPb
two astronauts example
Over time the challenge of identical twins eases slightly due to differences in environmental factors. But this does not fully remove the challenge.  

There are a few algorithmic approaches that can be applied for identical twins. One option could be a focus on Level III facial features like freckles and moles, which are indeed unique between identical twins. 

A more realistic approach for operating facial recognition technology in the presence of identical twins could entail administrative declaration of twin status. Knowing this information in advance allows for improved differentiation between twins while treating the rest of the population normally.

To a lesser extent, familial relationships also produce higher degrees of facial similarity, due to shared genetics. Even the “gold standard” of identification – DNA – faces similar limitations with twins and familial relationships. 

While facial appearance and DNA are driven by genetics, fingerprint friction ridge patterns and iris texture are not. Instead, they are formed environmentally (e.g., fingerprints are formed in the womb). As a result, fingerprint and iris algorithms do not share the same limitations with twins and relatives. 

 

Facial Spoofing

One of the biggest strengths for face recognition technology is the amount of publicly available data, which enables development of highly accurate algorithms – but this strength is a double-edged sword. 

The abundance of facial imagery on LinkedIn, Facebook, company websites, yearbook photos, and a wide range of other sources, creates substantial risk for identity fraud. Acquiring copies of face photos is exceptionally easy.

As we learned from Beyonce’s 2016 Super Bowl appearance, it’s essentially impossible to remove a photo from the Internet. Because facial appearance is our single most public piece of information, it means that there will be no solution in the form of removing facial images from everywhere they exist. 

Instead, to protect individual identities, significant effort must be invested in the development of spoof detection algorithms (also referred to as “liveness” checks or “presentation attack detection” algorithms).

Liveness checks are critical when face recognition is used for ID-verification through a mobile app or unattended customer service kiosks. As the accuracy of face recognition algorithms continues to increase at exponential rates, these unattended access control applications grow increasingly common. For other use cases, such as forensic identification or identity validation when a human operator is present (e.g., at a border crossing), liveness checks lack relevance. 

While fingerprint and iris biometric modalities can also be spoofed, acquiring samples to use in an attack is far more difficult, due to their private nature.

 

Decorative Cosmetics and Cosmetic Surgery

While facial appearance is largely determined by genetics, facial appearance can also transform through the use of makeup, hormones, and cosmetic surgeries.

Instead of cosmetics, early facial recognition research often focused on overt traits like facial hair. Over time, however, it has become increasingly clear that while factors like facial hair present a fairly insignificant challenge to facial recognition accuracy, cosmetics create more difficulty than originally understood. 

The fact that cultural use of cosmetics is predominantly by female populations has in turn resulted in minor, but statistically significant, differences in facial recognition accuracy between males and females. While some may attribute the difference in accuracy between genders to some inherent bias in facial recognition algorithms, it is becoming increasingly clear that the decrease in accuracy is instead due to the latent factor of cosmetics use by females.

 

Single Trait

While we have 10 fingers and two eyes, we only have one face. Thus, while it is more challenging to collect and manage samples from multiple fingers or irises, we can improve accuracy using this technique. With face recognition, this opportunity does not exist. 

 

Ground Truth Issues

Another challenge with face recognition technology is that authoritative databases often lack pristine identity labels. Whether due to fraud, human operator error, or other factors, prominent government databases all seem to have identity labels errors. These ground truth errors are not inherent for facial biometrics algorithms themselves, but instead stem from the long-term legacy use of facial imagery in databases prior to automated face recognition. 

Ground truth labeling errors can typically only cause an algorithm to measure higher error rates than actually exist (i.e., perform worse), so the error rates we capture in quantitative evaluations generally represent the upper bound for actual error rates on our algorithms. A portion of what’s measured as algorithm errors actually reflects data labeling errors. In other words, face recognition algorithms are more accurate than most benchmarks indicate.

Historically difficult to identify, these legacy errors in identity databases can reduce effectiveness of facial recognition algorithms. While ground truth errors are not easy to discover, automated face recognition algorithms have become astonishingly good at flagging potential errors for human review. However, this approach requires dedicated de-duplication processes.

Identity labels errors present significant challenges across all biometric traits, though fingerprint and iris present more unique challenges than compared to faces. While it may be easy to find ground truth errors in face recognition databases, finding them in fingerprint or iris databases is much more difficult. For potential facial identity error, the human brain is highly adept at comparing facial appearance making it easy to flag such inconsistencies when encountered. For fingerprint recognition and iris recognition, identifying errors is far more difficult, time-consuming, and expensive, often requiring expert analysis. 

Perhaps no method can reduce the incidence of ground truth errors more than the incorporation of multi-biometric modalities. Though many face systems support use cases where fingerprints are available, the use of both face and fingerprint biometrics significantly reduces the chances for fraud to exist in identity databases.

 

Summary

In the last two decades, the capabilities of automated face recognition technologies have undergone a stunning transformation. Once considered too inaccurate for use as a primary biometric trait, facial recognition is now the single most accurate biometric technology in the world. These exponential improvements will not likely slow anytime soon.

At the same time, every new technology comes with limitations. For facial recognition, that primarily includes identical twins and identity spoofing. 

The other two primary biometric traits –fingerprint and iris– also have advantages and limitations.

All factors considered, given the extreme ubiquity of facial appearance in our daily lives, and the astonishing accuracy of modern face recognition algorithms, minimal drawbacks exist for the technology. 

Most importantly, fusing face recognition with another mature biometric technology like fingerprint recognition can create a multi-modal biometric solution for nearly impervious unattended biometric identification.

Multi-modal biometrics are indeed the holy grail of identification. All biometric traits come with tradeoffs and weaknesses, but in scenarios where two or more disjoint traits can be measured together, the incident of failure or fraudulent access becomes exceptionally low. 

This article is a summary of a presentation provided at the National Institution of Standards (NIST) International Face Performance Conference (IFPC) on November 16th, 2022.

Put ROC multimodal biometrics to work for your business

Discover how our combination of face and fingerprint recognition can provide unparalleled accuracy for your identity verification and access control needs. Reach out now to upgrade your systems and safeguard your communities with only the best in multimodal biometric tech.

The post The Pros and Cons of Face Recognition appeared first on ROC.

]]>
ROC.ai Dominates Latest NIST Face Recognition Benchmarks https://roc.ai/2023/03/08/roc-ai-dominates-latest-nist-face-recognition-benchmarks/ Thu, 09 Mar 2023 00:10:48 +0000 https://roc.ai/?p=9013 The post ROC.ai Dominates Latest NIST Face Recognition Benchmarks appeared first on ROC.

]]>

Analysis of Face Recognition Vendor Test Results

Our latest release of the ROC SDK v2.4 delivers another substantial set of improvements, demonstrated by its exceptional performance in the February 2nd, 2023 National Institute of Standards and Technology (NIST) Face Recognition Vendor Test (FRVT) Ongoing benchmark. 

While our organization has maintained a steady cadence of enhancements to our face recognition algorithm every 4 to 6 months since founding in 2015, the last 18 months show a clear surge in the pace of our improvements.

Labeled as “rankone-014” in the latest FRVT Ongoing report, the ROC SDK v2.4 face recognition algorithm outperformed hundreds of global competitors across numerous critical benchmarks, achieving: 

  • #1 global leader in combined accuracy and efficiency
  • #7 of 338 global solution providers in lowest Average Error Rate
  • #1 of all key marketplace and U.S. competitors in Average Error Rate
  • #6 of 338 global solution providers in lowest Visa Border dataset error rate
  • #1 lowest Visa Border dataset error rate of all key marketplace and U.S. competitors

Specific performance metrics from ROC SDK v2.4 in the NIST FRVT Ongoing 2/2/23 report, labeled as “rankone-014” in Tables 8 to 29, are as as follows:

ROC SDK v2.4 Face Recognition Algorithm Accuracy and Efficiency Metrics

These error rates represent a 1.3x decrease as compared to our previous ROC SDK v2.3 (rankone-013 in FRVT). Accuracy rates are now consistently in excess of 99.5% accuracy / True Match Rates, while in many cases operating at False Match Rates of 1 in 1,000,000 (1E-6). 

As shown above, the Average Error Rate of ROC v2.4 now measures at 0.0061. This is simply computed from Tables 19 to 29 in the FRVT Ongoing report by measuring the arithmetic mean of all eight error rates liste

When comparing our Average Error Rate to other vendors listed, ROC.ai now ranks 7th of 338 global algorithm developers:

Face Recognition Average Error Rates: Global Top 50
Bar chart of the 50 global face recognition developers with the lowest Average Error Rate in FRVT Ongoing 2/2/2023. Rank One Computing is highlighted at 7th out of 338 total developers benchmarked.

When looking at key marketplace competitors, ROC.ai delivers the lowest average error rate: 

Face Recognition Average Error Rates: Market Competitors
Rank One Computing shown with the lowest average error rate out of ten market competitors in FRVT Ongoing 2/2/2023.

In addition to achieving best across-the-board accuracy of all key market competitors in FRVT Ongoing, ROC.ai also placed 6th globally out of 318 vendors in the Border dataset. When examining key competitors on the Visa Border dataset, ROC SDK easily had the lowest error rate:

Visa Border Face Recognition Error Rates: Market Competitors
Bar chart of ten market competitors with the lowest Visa Border error rate in FRVT Ongoing 2/2/2023. Rank One Computing is shown with the lowest error rate.

Ranking 7th globally in accuracy, the ROC SDK trails the global leader in each category by only only a fractions of a percentage points:

Comparison of ROC v2.4 vs. Top Global Vendor Per Category

While Chinese companies are leading 7 of the 8 categories, ROC SDK is often less than 0.1% away from the global leader in each category. 

Efficiency Analysis

While ROC.ai is a global leader in accuracy, we are also truly one-of-a-kind in combined accuracy and algorithmic efficiency. 

The four summary statistics for face recognition algorithm efficiency are provided in Tables 8 to 18 in the FRVT Ongoing report are: 

  • Template size – the amount of bytes need to represent a face image
  • Comparison time – the amount of time it takes to compare two templates and generate a threshold.
  • Template generation time – the amount of time to process a face and produce a comparable template
  • Binary size – the amount of memory / RAM needed by a device to run the algorithm

More information about these efficiency metrics can be found in our previous article, Procuring a Face Recognition Algorithm: Efficiency Considerations, as well as Hardware Considerations when Architecting a Face Recognition System.

For each of these four efficiency metrics, the NIST FRVT Ongoing report ranks each of the algorithms amongst all 478 algorithms (from 338 developers), with the best performing algorithm ranking #1. While the top ranks are often from highly inaccurate algorithms, ROC is unique in both ranking as both one of the most efficient algorithms and one of the most accurate algorithms. 

The ranking of each of ROC SDK’s algorithm efficiency metrics is as follows: 

ROC SDK v2.4 Face Recognition Algorithm Efficiency Metrics and Rankings

Out of the 478 algorithms, ROC SDK v2.4 has the 13th best average efficiency ranking. Our previous versions were also included in the benchmark test – ROC SDK v2.2 has the 12th best efficiency ranking, and ROC SDK v2.0 has the 7th best efficiency ranking). When examining the Average Error Rate across the top 20 most efficient algorithms, it is fairly stunning how much lower the error rate is with ROC compared to the other highly efficient algorithms: 

Face Recognition Efficiency and Error Rates: Global Top 20 by Efficiency Rank

ROC SDK provides error rates that are 2x to 50x lower than the other highly efficient algorithms. Only two of the top 20 most efficient vendors perform within 4x of ROC’s error rate, and most resulted in a more than 10x higher error rate.

When examining marketplace competitors, our efficiency ranking is substantially better than the top 10 competitors with lowest error rates: 

Efficiency and Error Rates: Market Competitors by Error Rate

ROC SDK combines the lowest error rate of all competitors, with unbeatable efficiency compared to  marketplace competitors. 

Finally, when examining the top 20 globals vendors with the lowest error rate, Rank One Computing emerges as being unparalleled in average efficiency ranking:

Efficiency and Error Rates: Global Top 20 Vendors by Error Rate

All together, ROC.ai is the unquestioned global leader in combined accuracy and efficiency.

FRVT 1:N Performance

In addition to ranking 7th globally in the FRVT Ongoing report, the ROC SDK v2.4 also achieved standout performance in the Feb 10th, 2023 NIST FRVT 1:N Identification report. The 1:N report measures the accuracy of searching large databases, which is an operational use-case with a long and historic legacy. 

Specific achievements of ROC.ai in the 2/10/2023 FRVT 1:N report include:

  • Top-10 globally in Rank-1 hit-rate in all investigative search accuracy results (64K image dataset up to 12M image dataset)
  • #8 of 388 algorithms in investigation frontal mugshot ranking 
  • #6 of 250 algorithms in investigation mugshot webcam ranking 
  • #9 of 277 algorithms in immigration visa border ranking 
  • #5 of 222 algorithms in immigration visa kiosk ranking
  • Rank-1 hit rate of 99.87% searching a database of 12M images

Demographic Biases in Face Recognition

Another key achievement of the ROC SDK v2.4 was further improvements in normalizing error rates and match thresholds across key demographic groups. While certain demographic plots for ROC SDK v2.4 / ‘rankone_014’ were not yet published in the 2/2/2023 FRVT Ongoing report, our organization still notably ranked 7th out of 406 algorithms in lowest False Match Rate (FMR) ratio between West Africa and Eastern Europe, according to the FRVT Demographic Effects in Face Recognition online leaderboard. We also ranked 7th globally for lowest False Non-Match Rate (FNMR) overall. 

Further examining the vendor scorecard for ROC SDK v2.4, the FNMR rates across a wide range of countries was quite stable: 

The differences in False Match Rate (FMR) between different countries and age groups was also highly stable, where the following results show the factor of difference in FMR at global threshold that corresponds to a FMR of 0.00003: 

These results demonstrate the stability of the ROC SDK across a wide range of demographic cohorts.

Face Recognition Accuracy Improvements Over Time

Finally, we will examine the improvement that ROC SDK has delivered over time to our partners and customers. 

From the FRVT scorecard, the following plot summarizes our error rate reductions over the last 5+ years: 

Over this time the FNMR has been reduced by over 100x. In just the last 18 months, ROC has released four new FR algorithms (in ROC SDK v1.26, v2.0, v2.2, and v2.4). Each of these releases has delivered an average error rate reduction on the NIST FRVT Ongoing benchmark between 1.3x and 1.4x. In total, this has results in a 3.4x error rate reduction over this period of time with roughly a 5x reduction on the Visa datasets: 

These improvements are not slowing down. ROC.ai is currently working on our next face recognition algorithm that will be released in the coming months. 

Summary

This comprehensive analysis of latest NIST results demonstrates a clear pattern of superior performance. ROC.ai dominates Western countries in face recognition accuracy and efficiency. We rank first in lowest Average Error Rate among all marketplace competitors, and in efficiency compared to all top-tier algorithms. In addition to being unparalleled in combined accuracy and efficiency, ROC.ai also delivers unparalleled customer support and straightforward business practices. Contracting mechanisms like Evergreen Licensing ensure that our customers and partners always have access to all the cutting-edge advancement being delivered by our R&D. 

Our differentiation does not end with accuracy, efficiency, and customer satisfaction. ROC.ai has established a rock-solid reputation as the most trustworthy developer of facial recognition algorithms in the world. We are employee-owned and financially-independent. Our technology is developed fully in-house in the U.S.A. We do not answer to the whims of investors who may have short-term goals that do not align with our customers, partners, employees. Instead, we are building a collaborative long-term roadmap to deliver faster, safer, more reliable computer vision solutions that can dutifully serve the greater needs of the world at-large.

The reasons are endless to begin integrating ROC SDK v2.4 face recognition algorithms into your organization. ROC.ai also offers top-tier algorithms for fingerprint detection, object detection, license plate recognition (LPR), optical character recognition (OCR), and more.

Reach out to team up with the world’s most trusted developer of face recognition solutions.

Get started

The post ROC.ai Dominates Latest NIST Face Recognition Benchmarks appeared first on ROC.

]]>
ROC AI’s Fingerprint Algorithms Achieve Best-in-Class Accuracy https://roc.ai/2023/02/15/roc-ais-fingerprint-algorithms-achieve-best-in-class-accuracy/ Wed, 15 Feb 2023 19:33:29 +0000 https://roc.ai/?p=8754 Fingerprint capabilities delivered in ROC SDK v2.4 lead world in accuracy and efficiency, as demonstrated by the statistical results presented in this article.

The post ROC AI’s Fingerprint Algorithms Achieve Best-in-Class Accuracy appeared first on ROC.

]]>
ROC SDK v2.4 Fingerprint Analysis

The following report documents ROC AI’s performance in the National Institute of Standards and Technology (NIST) Proprietary Fingerprint Template (PFT) III benchmark. Competitor plots were generated on 6 February 2023.

With the exponential explosion in the capabilities of face recognition algorithms, focus has in many ways shifted from what has been the most authoritative biometric trait the last several decades: fingerprints. 

While facial appearance has been the default biometric trait throughout all of human existence, this role has always been through innate, subconscious cognitive activity. When instead examining the systematic use of biometric traits for identification purposes, the history of fingerprint recognition is substantially larger than any other trait. And, from an automated biometric perspective, fingerprint recognition has long since held the distinction as the most trusted biometric.

Fingerprint recognition systems are used by nearly every country in the world for a range of critical identification infrastructure tasks. And while they provide extreme trust, they also contend with significant computational efficiency bottlenecks. Specifically, the comparison speed for fingerprint algorithms has historically been extremely slow. This is due to fingerprint recognition being treated as a point-set matching problem with minutiae location, orientation, and type being used as the point sets. 

As the pattern recognition technology fully progresses in the modern era of deep learning, should fingerprint recognition systems still be bottlenecked by legacy constraints? 

The answer is no, as fully demonstrated with the release of ROC AI’s new fingerprint recognition capabilities delivered in the ROC SDK v2.4.   

ROC AI’s fingerprint solutions leverage the same trade secrets that have enabled ROC SDK face recognition capabilities to lead the industry in combined accuracy and efficiency for the last five years. The end result is a fingerprint algorithm that:

  • Delivers best-in-class accuracy; and
  • Operates at an efficiency range that resets expectations on the scalability of fingerprint systems.

The remainder of this article provides dense statistics from the National Institute of Standards and Technology (NIST) Proprietary Fingerprint Template (PFT) III benchmark

Performance metrics for the ROC SDK v2.4 fingerprint algorithm, per NIST’s scorecard, are as follows: 

In terms of performance relative to all other vendors who have submitted to the NIST PFT benchmark, ROC AI is the number one performer on the Nail-to-Nail benchmarks, ranking #1 in 8 of the 12 sensors, #2 in the other four sensors, and #1 in mean error rate across all sensors:

For the remaining PFT test sets, ROC AI was #2 in three of the four sets and top three in all sets:

In terms of mean error rate across all PFT III test sets (all Nail-to-Nail sensors, AZ, LA, Port of Entry, and US VISIT), ROC AI has the lowest mean error rate of all vendors:

Of course, similar to all ROC AI algorithms, it is not just accuracy but also efficiency that sets ROC AI apart. In fingerprint this is also the case:

The ROC SDK v2.4 fingerprint algorithm uses a substantially smaller template than any other vendor in NIST PFT, and has the fastest comparison comparison speeds (more than 1000x faster than many key competitors). 

The combination of lowest error rates and best computational efficiency truly puts ROC AI in a class of it’s own in the fingerprint industry: 

Indeed, no other vendor can match this combination. In addition to accuracy and efficiency distinctions, the ROC AI fingerprint algorithms, like all of our algorithms, are developed entirely “in-house”, by ROC AI employees, in the United States of America. 

While the current ROC SDK v2.4 fingerprint algorithm catapults ROC AI to the top of the fingerprint capabilities in the world, it is important to remember that this is in fact the first fingerprint algorithm released by ROC AI. Similar to the pace of improvements delivered in face recognition, multiple new releases for fingerprint recognition will be delivered by ROC AI in 2023 and beyond. And, ROC AI will continue to provide our customer friendly “evergreen licensing” terms to our partners and customers. 

Reach out now to start your journey toward integrating the best-in-class ROC AI fingerprint capabilities!

 

The post ROC AI’s Fingerprint Algorithms Achieve Best-in-Class Accuracy appeared first on ROC.

]]>
ROC SDK v2.3 Delivers More Algorithm Improvements https://roc.ai/2022/11/10/roc-sdk-v2-3-2/ Fri, 11 Nov 2022 03:53:28 +0000 https://roc.ai/?p=8264 The post ROC SDK v2.3 Delivers More Algorithm Improvements appeared first on ROC.

]]>
ROC SDK version 2.3 continues to demonstrate the power of our AI/ML computer development advancements with significant improvements to the following algorithms: 

Liveness / Presentation Attack Detection (PAD)

Perhaps the most consequential improvement with v2.3 is a major reduction in error rates to ROC’s patented, single-frame, passive liveness algorithm. Rank One’s investment in liveness capabilities has yielded substantial improvements over the last year, building on the recent improvements in our previous v2.2 release.

The liveness solution currently ships with two different suggested operational thresholds: Security and Convenience. Convenience mode is primarily relevant to continuous authentication applications that need to be as frictionless as possible. Security mode is primarily relevant to single authentications related to sensitive systems and assets, such as identity proofing. 

The following improvements have been delivered to the ROC SDK v2.3 Liveness solution:

Liveness error rate comparison between ROC SDK v2.2 and v2.3

The provided error rates are measured on a large corpus of images that span all the Level A spoofing attacks as defined by the FIDO Alliance and in accordance with the ISO 30107-3 testing standard. The Genuine Reject Rate, which is referred to as the Bona Fide Presentation Classification Error Rate (BPCER) in the ISO standard, is when a genuine user’s presentation to a camera is incorrectly flagged as a potential spoof presentation. The Spoof Accept Rate, which is referred to as the Attack Presentation Classification Error Rate (APCER) in the ISO standard, is when a spoof presentation is incorrectly accepted as a genuine user presentation.

As shown in the above results, the ROC Liveness solution now provides the ability to severely limit spoof attacks while keeping the acceptance of genuine samples within the standards-compliant range. ROC will be eagerly submitting to the forthcoming NIST FRVT PAD benchmark. Through these improvements ROC will become certified as fully compliant with all applicable standards and industry best practices for facial biometric presentation attack detection.

Beyond single frame passive liveness, ROC further provides a suite of capabilities to support digital identity verification. This includes a wide range of facial analytics in support of ICAO and ISO standards. ROC offers easy to integrate “LiveScan” capabilities that allow ROC partners to easily capture a single, standards-compliant facial image for a user presenting themselves to a camera, with or without an active internet connection. 

Another round of enhancements to ROC Liveness will be on the way at the start of 2023. 

Tattoo Recognition

Rank One is a leading provider of tattoo recognition technology, with our algorithms deployed domestically and internationally within various law enforcement agencies. This capability is primarily used by law enforcement agencies with large databases of tattoo images captured at arrest booking to perform tasks such as identifying deceased victims who have little information available for identification aside from their tattoos. 

In the latest version of ROC’s tattoo algorithm, a major improvement enables the ability to accurately detect, localize, and represent tattoos images in a compact feature vector representation. Through the implementation of best practices in deeply convolved neural network representations of localized tattoo regions, ROC has now achieved drastically higher accuracy.

When evaluating 1:1 comparison accuracy for the ROC tattoo algorithm, which allows for easy generalization to the true 1:N use-case, the ROC tattoo algorithm achieves robust performance on this challenging problem: 

ROC v2.3 Tattoo Recognition Accuracy Results

Such recognition accuracies are quite powerful, especially when compared to the last published NIST Tatt-E report. While NIST Tatt-E is not currently accepting new submissions, accuracy comparisons of the top performing submission in the NIST Tatt-E report to ROC Tattoo v2.3 confirm that the ROC algorithm is significantly more accurate.

NIST Tatt-E primarily measured accuracy as Rank-10 retrieval rate on a gallery of 100,000 images. In this manner, the most accurate solution achieved a hit rate of 72.1% in that report. While not directly comparable, True Accept Rate at a False Accept Rate of 1 in 100,000 (0.001%) would serve as a lower bound for Rank-1 accuracy on a gallery of size 100k in cases where there is only one sample per identity.

By comparison, when the ROC algorithm is measured at a False Accept Rate of 0.001%, a True Accept Rate of 90.8% is achieved. Thus, increasing the challenge from being a Rank-10 match to a Rank-1 match, ROC (90.8%) still outperforms the industry leading solution (72.1%) by a wide margin. Such a clear separation in recognition accuracy demonstrates that the new ROC Tattoo algorithm is likely the most accurate solution in the world (and by a wide margin)

Accuracy is not the only thing setting the ROC tattoo algorithm apart from participants in the NIST-C report. 

ROC v2.3 Tattoo Recognition Accuracy Metrics on a single a x64 CPU Core

As shown in the above table, the ROC Tattoo algorithm is incredibly fast and efficient. By comparison, in the NIST Tatt-E report the most accurate algorithm required 75.8 seconds to conduct a single search of a 100k image dataset. ROC can do this in less than 350 milliseconds. In other words, the ROC tattoo algorithm is not just the most accurate tattoo algorithm, it is also over 100x faster than the previously most accurate solution! Even the fastest solution submitted to NIST required 2.0 seconds to conduct the same search despite being far less accurate than other algorithms in that report.

In addition to the quantitative results of this new algorithm, the algorithm’s qualitative performance is arguably more impressive. While privacy reasons prevent us from showing operational tattoo images, top retrieval candidates generally have highly visual similarity to the probe candidate images. Such results make it immediately clear that this next-gen tattoo recognition capability will be a game-changer for forensic investigators in terms of the ability to find the same or highly similar tattoos to one in question. 

Facial Analytics

As opposed to face recognition which is purely based on the identity of a person contained in an image, facial analytics provide ancillary information regarding the presented face.  

Rank One has been an industry leader for many years in automated facial analytics, powering use-cases ranging from age verification, to LiveScan face acquisition for passenger travel, retail analytics, and many more. 

To address the rising demand for ROC Facial Analytics for deployment across a wide range of hardware architectures and software systems, the ROC SDK v2.3 delivers a complete overhaul of our analytics algorithms. All facial analytics can now be generated without first computing a facial recognition template. Through this change our full suite of facial analytics algorithms can be extracted in roughly 50ms total on a single CPU thread. These analytics include: 

  • Age
  • Gender
  • Geographic Origin
  • Emotion
  • Facial pose
  • Glasses
  • Mask
  • Eyes Visible
  • Occlusion
  • Facial Hair

In addition to the speed improvements from this new facial analytics method, it also delivers accuracy improvements to a majority of these different analytics methods. 

Finally, we added one new facial analytic feature to this release: the ability to remove the background from a facial photograph. In turn, the background of the photograph can be set to a consistent color in order to adhere to various compliance standards such as  ISO/IEC 19794-5 and ICAO 9303.

License Plate Detection

The final algorithmic improvement in ROC SDK v2.3 is a significant increase in accuracy for our license plate detection algorithm. This license plate detector is primarily used with ROC’s overall License Plate Recognition (LPR) solution.

Coming on the heels of the LPR improvements in v2.2, ROC is quickly developing broad capabilities in a range of use-cases for LPR technology.

Contact us today to learn more about ROC SDK v2.3! 

The post ROC SDK v2.3 Delivers More Algorithm Improvements appeared first on ROC.

]]>
Rank One Places #1 in the World in Latest NIST FRVT Quality Benchmark https://roc.ai/2022/09/07/rank-one-places-1-in-the-world-in-latest-nist-frvt-quality-benchmark/ Wed, 07 Sep 2022 13:59:47 +0000 https://roc.ai/?p=8061 From the FRVT Quality website leaderboard Rank One is the best overall algorithm at filtering out low quality samples while maintaining a low False Non-Match Rate (FNMR) for the face recognition algorithm

The post Rank One Places #1 in the World in Latest NIST FRVT Quality Benchmark appeared first on ROC.

]]>
There is an old adage in systems engineering: garbage in, garbage out.  

No matter the source of the data used by an analytical system, if it is not captured in a manner conformant with how the algorithms and system were designed then it will not yield an ideal outcome. 

Across the vast range of facial recognition use-cases it is critical to prevent “garbage” from undermining the effectiveness of the system. Whether it is border screening applications, online selfie enrollment for banking and enterprise services, law enforcement use cases, or just about any other use of face recognition technology, there is a substantial benefit for preventing face images that are of insufficient quality from ever being processed. 

The way to prevent this “garbage” from being processed by a system is to flag bad images at the time of capture, which is the role of automated facial quality algorithms. Factors that may make an image lower quality include heavy blur, occlusions, low resolution, or extreme pose angles. 

Effective facial quality algorithms are so important to the deployment of face recognition systems that the National Institute of Standards and Technology (NIST) Face Recognition Vendor Test (FRVT) has a dedicated benchmark on “Face Image Quality Assessment”.

ROC delivers on quality

Rank One recently delivered to its customers a substantial improvement to its quality algorithm. This new quality algorithm as also submitted to NIST FRVT and the results are in: Rank One’s latest quality algorithm is #1 in the world! 

From the FRVT Quality website leaderboard Rank One is the best overall algorithm at filtering out low quality samples while maintaining a low False Non-Match Rate (FNMR) for the face recognition algorithm:  

Figure 1: Excerpt from the NIST Quality Website Leaderboard

Source: NIST FRVT Quality Summarization and Analysis  https://pages.nist.gov/frvt/html/frvt_quality.html

Per the NIST benchmark, by filtering out images with the lowest 1% in quality the FNMR is reduced from 0.01 to 0.0059. No other vendor is capable of achieving such a reduction in FNMR. 

This new quality algorithm, which is available in the latest version of the ROC SDK,  provides Rank One’s customers with a simple and effective method for rejecting non-conformant imagery from the system. In addition to this single automated quality metric, the ROC SDK also provides the ability to ensure facial imagery conforms with other specific image quality checks such as the ICAO standard

Rank One’s achievement in the NIST FRVT Quality benchmarked coupled with Rank One’s recent standout performance in the NIST FRVT 1:1 and 1:N benchmark further underscores what Rank One’s customer already know: there is not a more trusted provider of accurate, efficient, and easy-to-use face recognition technology than Rank One Computing.

About Rank One

Founded in 2015, ROC is an employee-owned company with headquarters in Denver, CO and offices in Morgantown, WV that develops its software entirely in-house and in the U.S.A. ROC delivers top performing face recognition and computer vision algorithms with a no-nonsense business approach and deep commitment to best practices in software engineering and pattern recognition algorithm design. We have initiated – and continue to lead – the charge to develop responsible AI by establishing the FR industry’s first code of ethics that governs our development and deployment of AI/ML algorithms and software in both commercial and government applications.

The post Rank One Places #1 in the World in Latest NIST FRVT Quality Benchmark appeared first on ROC.

]]>
Rank One Computing Ranks #1 in Latest NIST FRVT for Combined Accuracy and Efficiency https://roc.ai/2022/08/10/rank-one-computing-ranks-1-in-latest-nist-frvt-for-combined-accuracy-and-efficiency/ Thu, 11 Aug 2022 04:00:06 +0000 https://roc.ai/?p=7291 ROC’s SDK v2.2 is now the single-highest performing vendor of all U.S. FR algorithm providers in the latest National Institute of Standards and Technology (NIST) Face Recognition Vendor Test (FRVT). Specifically, v2.2 outranks all others with both top tier accuracy and efficiency in both the 1:1 and 1:N series.

The post Rank One Computing Ranks #1 in Latest NIST FRVT for Combined Accuracy and Efficiency appeared first on ROC.

]]>

ROC’s SDK v2.2 is now the single-highest performing vendor of all U.S. FR algorithm providers in the latest National Institute of Standards and Technology (NIST) Face Recognition Vendor Test (FRVT). Specifically, v2.2 outranks all others with both top tier accuracy and efficiency in both the 1:1 and 1:N series. At 2-5x faster and 10-20x more efficient, v2.2 packs a punch of impressive computer vision developments: 5x improvement in liveness fraud detection, expanded PersonID searching and brand new LPR functionality. These performance improvements will directly enhance performance of ROC Watch and all ROC applications. Contact us to ensure you have up to date versions of our top-ranked SDK.

True to our company culture, ROC continues to deliver the most efficient, accurate, and trusted computer vision algorithms available on the market today.  I’m incredibly proud of the work we have done in a few short months since releasing our next generation AI/ML-powered version 2.0. In that release, we included some exciting new computer vision capabilities including Liveness fraud prevention, Vehicle and Object Detection and License Plate Recognition (LPR) based on Optical Character Recognition (OCR). Read more about v2.0 in our previous blog.

With the v2.2 release, our AI/ML developers have made impressive gains that deliver powerful improvements to Liveness fraud prevention, LPR, and Person ID in addition to ground-breaking accuracy performance for Face Recognition – all in a single, efficient SDK packet.

Face Recognition  Improvements

While accuracy metrics are the primary currency of FR vendors, integrators of FR systems know that real-world performance of these algorithms requires that they are deployable in real-world settings, such as high throughput video processing, large-scale enterprise search systems, or on-edge mobile use-cases. For this reason, ROC continues to prioritize speed and efficiency in parallel to our continued drive for top accuracy performance.  And with v2.2, we have continued to establish ourselves as the industry leader.

Just as smart phones require chips that are small, efficient, and fast – FR applications require algorithms that are efficient, fast, and accurate.  With ROC SDK v2.2, you don’t have to choose between accuracy and efficiency.

As shown in Figure 1 below from the 07/28/2022 NIST FRVT Ongoing report, Rank One is, by a wide margin, able to deliver the best combination of accuracy and efficiency when measuring the mean FRVT rankings in all eight accuracy benchmarks (Tables 18 to 27: Visa MC, Visa, Mugshot, Mugshot 12+Yrs, VisaBorder, Border 10-6, Border 10-5, Wild) and all four efficiency benchmarks (Tables 8 to 17: template generation speed, template size, binary size, and comparison speed.)

Figure 1: Mean Accuracy and Efficiency Ranking across 28 Commercial FR Vendors

When focusing specifically on accuracy metrics, v2.2 continues ROC’s impressive climb in rankings.  For example, as shown in Figure 2 below, ROC SDKv2.2 is ranked as the #1 most accurate U.S. vendor on Border crossing data (1E-5). v2.2 also ranked #6 globally of 440 algorithms total with all of the top 5 developed in China, Russia, and Korea.

Figure 2: Error Rates for on Border Data (1E-5) for U.S. Vendors

And more broadly, ROC v2.2 achieved better than 99.5% genuine match rate on all Border, Visa, and Mugshot datasets. Figure 3 demonstrates this as well as the dramatic 1.4x improvements that ROC has made in just the last six months. 

Figure 3: ROC SDK v2.2 Error rate improvements over v2.0

Liveness Fraud Prevention

As defined by the Presentation Attack Detection (PAD) standards outlined in ISO/IEC 30107-3, Liveness detection seeks to combat the following attempts to make a fraudulent FR match:

  • 2D static attacks: high-definition face pictures on flat paper or simple flat paper masks with holes are presented to the FR matcher
  • 2D dynamic hacks: multiple images (2D photos or 3D avatars) are played in sequence on a 2D screen
  • 3D static attacks: impersonators use 3D prints, wax heads, or sculptures
  • 3D dynamic attacks – impersonators use sophisticated masks to imitate liveness.

v2.2 achieved massive reductions in PAD error rates. Bona Fide Presentation Classification Error Rate (BPCER) reduced by 7x when operating at our “High Security” threshold and 5x when operating at our “Low Security” threshold. Attack Presentation Classification Error Rate (APCER) held constant at these operating thresholds.

New PersonID Functionality

PersonID refers to the capability to recognize persons through morphology (shapes and colors). v2.2 adds the ability to use clothing color semantics to query a video feed for persons wearing such clothing.  For example, using v2.2 an analyst query can be performed to “Find all images that include a person with a yellow shirt and gray pants.”  Additionally, the ability to compare the visual similarity of two persons using computer vision representations is still supported in the ROC SDK. Together, this PersonID functionality supports both real-time monitoring and forensic investigation use-cases.   

Expanded License Plate Recognition

Building on the v2.0 debut of LPR, v2.2 tackles specific challenges associated with recognizing the many different license plate formats and layouts issued by states, municipalities, and agencies.  This release delivers major improvements in the ability to detect the U.S. state of origin for license plates and to accurately recognize license plate digits using Optical Character Recognition (OCR).  To achieve these results, ROC developed tailored training data for its AI/ML algorithms to improve performance in operationally relevant conditions.

About Rank One

Founded in 2015, ROC is an employee-owned company with headquarters in Denver, CO and offices in Morgantown, WV that develops its software entirely in-house and in the U.S.A. ROC delivers top performing face recognition and computer vision algorithms with a no-nonsense business approach and deep commitment to best practices in software engineering and pattern recognition algorithm design. We have initiated – and continue to lead – the charge to develop responsible AI by establishing the FR industry’s first code of ethics that governs our development and deployment of AI/ML algorithms and software in both commercial and government applications.

The post Rank One Computing Ranks #1 in Latest NIST FRVT for Combined Accuracy and Efficiency appeared first on ROC.

]]>
ROC SDK Version 2.0: A New Generation of Visual Analytics Algorithms  https://roc.ai/2022/02/03/roc-sdk-version-2-0-a-new-generation-of-visual-analytics-algorithms/ Fri, 04 Feb 2022 02:25:14 +0000 https://roc.ai/?p=5926 ROC SDK version 2.0 delivers both substantial face recognition improvements and a completely new feature set of computer vision and machine learning capabilities. Specifically, the SDK now supports detection of a wide range of objects and text recognition on a wide variety of challenging images (e.g., license plates in the wild). 

The post ROC SDK Version 2.0: A New Generation of Visual Analytics Algorithms  appeared first on ROC.

]]>
Rank One Computing was founded in 2015 with a mission to deliver top performing face recognition and computer vision algorithms with a no-nonsense business approach and deep commitment to best practices in software engineering and pattern recognition algorithm design.

Over the course of the last seven years Rank One has raised the industry-wide bar on what is possible with a software development kit. The ROC SDK has demonstrated that top-tier face recognition accuracy can be delivered alongside algorithm efficiency metrics that are often 10x better than industry peers. It has further demonstrated that API’s for an SDK can be lightweight but highly effective and easy to integrate. Rank One’s technical support has consistently separated us from other solutions, because our lead engineers provide rapid responses to technical inquiries. ROC SDK gives trusted biometrics a new meaning since Rank One is an employee-owned company that develops its software entirely in-house and in the U.S.A. Finally, Rank One has led the way by establishing the facial recognition industry’s first code of ethics to govern the use of our algorithms and software in both commercial and government applications.

Building on this strong foundation, Rank One announces a new chapter in its algorithmic offerings with the release of the ROC SDK version 2.0.

ROC SDK version 2.0 delivers both substantial face recognition improvements and a completely new feature set of computer vision and machine learning capabilities. Specifically, the SDK now supports detection of a wide range of objects and text recognition on a wide variety of challenging images (e.g., license plates in the wild).

Introducing ROC SDK 2.0
The significant improvements delivered with ROC 2.0 include:

Face Recognition Algorithms
The following analysis is based on the National Institute of Standards and Technology (NIST) Face Recognition Vendor Test (FRVT) report released on January 20th, 2022. All statistics provided are directly as listed in the report, which is generally from Tables 7 to 24.
Algorithm Improvements
On the heels of v1.26 that delivered substantial accuracy improvements just five months ago, ROC SDK Version 2.0 delivers some of the largest face recognition accuracy improvements ever to the ROC SDK.. All together ROC has cut error rates in half over the last four months as demonstrated by the following reduction in error rates in the NIST FRVT Ongoing benchmark:
Accuracy Compared to Industry Peers
With these substantial error rate reductions, ROC SDK v2.0 – the entirely U.S. owned and developed facial recognition SDK – has basically reached performance parity with the highest ranked FRVT performers which are all currently of Chinese origin:
ROC 2.0 consistently ranks as one of the top-performing algorithms across a wide range of datasets and operating conditions measured by NIST FRVT. The chart below further demonstrates the top-tier accuracy delivered by ROC 2.0 when compared against the median performing algorithms. On the y-axis is False Non-Match Rate, shown on a scale of 0% to 4%, whichideally is minimal (lower is better):
Algorithmic Efficiency
The ROC SDK is the effective industry leader in algorithm efficiency while also providing top-tier accuracy. Truly no other vendor can claim this distinction. As compared to the median FRVT algorithm, the ROC SDK is substantially more efficient across the board:
Other industry algorithms, particularly high accuracy algorithms, require substantial hardware resources compared to the ROC SDK. The implication of top-tier accuracy alongside top-tier efficiency is that integrators and customers can save substantial amounts of money on hardware costs. In certain cases, whether it is a high quantity of on-edge processors, or a centralized processing environment, the savings in hardware costs when using the ROC SDK may cover the cost of licensing the SDK. Thus, taking into account the hardware requirements of industry peers, the ROC SDK can effectively become free. This hardware processing cost advantage scales not only to on-premise deployments but also to cloud deployments on virtual servers or serverless environments which typically charge by CPU hours, RAM, and storage.
Algorithmic Bias
In terms of algorithmic bias, Rank One continues to exhibit minimal differences in error rate between racial and gender cohorts:
The above chart shows the performance variation among white, black, female and male sub-demographics and specifically, the tradeoff between false positive and false negative error rates for each sub-demographic. While the error rates are exceptionally low across all four race / gender cohorts, the lowest overall error rate is with the Black Male cohort (the red, dotted line), particularly at the operational thresholds of 10-4 (1 false match in 10,000 comparisons) and 10-5 (1 false match in 100,000 comparisons). In general the Black Female cohort is the second lowest error rate of all four cohorts (the red, solid line). Such exceptional performance across racial cohorts is counter to the misinformation campaigns that have been propagated by prominent media outlets. But, as continues to be exhibited: top-tier face recognition algorithms are highly accurate on all races.
Summary of Current ROC 2.0 Performance Metrics
The following table provides the summary of ROC’s FRVT performance alongside the median algorithm performance for reference:
Long Term Improvements
Finally, it is important to recognize the orders of magnitude with which the ROC SDK has improved over the past several years. The following plot provided by NIST shows this error rate reduction:
The y-axis is plotted on a logarithmic scale because the improvements are so substantial. Indeed over the course of a few years error rates have been reduced 50x!

These improvements are stunning given how successful the algorithms were in years past. And while such improvements are often being delivered by industry providers, Rank One continues to stand out with the practice of Evergreen Licensing, which offers a simple approach for ROC customers and partners to receive continued access to such ground breaking accuracy improvements.

While the recent improvements to the ROC SDK have reaffirmed its status as the industry leader in top-tier accuracy and efficiency, the innovation train continues to move at a rapid pace via Rank One’s vigorous R&D initiatives. In Spring 2022, we expect to release a new ROC SDK version with yet another powerful set of improvements to the depth of our FR capabilities and breadth of our computer vision capabilities.

Facial Analytics
While Rank One is typically recognized for its excellence in automated face recognition, there are many facial analytics tasks that differ from face recognition itself, but at the same time are highly complementary. The ROC SDK provides some of the best facial analytic tools available, including facial liveness (presentation attack detection), ICAO compliance checks, demographic estimation, occlusion detection, encrypted matching, and facial quality estimation.

With the release of ROC 2.0, substantial improvements are being delivered to this portfolio of facial analytics.

Liveness
Liveness validation is an automated check to ensure that an image is an actual live presentation of the subject, as opposed to a photograph of that person being held up to the camera (presentation attack). Rank One has a patented micro-texture approach to liveness validation that has been used operationally by ROC customers for several years. And, with ROC SDK 2.0 a new liveness algorithm is being delivered that has significant accuracy improvements. Rank One will deliver more liveness algorithm improvements in 2022, and is in the process of obtaining various certifications for these algorithms, including certification from iBeta for biometric presentation attack detection (PAD) per ISO/IEC 30107-3. 
ICAO Compliance
As passenger travel increasingly relies on face recognition to perform identity verification, a wide range of laws and regulations require that facial photographs of passengers are captured in a manner that complies with the ICAO Portrait Quality for Machine Readable Travel Documents standard. For example, in the European Union all airline passengers must provide an ICAO compliant facial scan prior to boarding a flight in accordance with the Schengen Area Entry/Exit System.

Beyond travel applications, capturing a high quality enrollment photo is a prerequisite for digital identity solutions, access control solutions, and broadly any use case with a database of authorized users. The ROC SDK version 2.0 automated ICAO/ANSI-NIST quality metrics can enable any use case to automatically capture a compliant image from a streaming video source, and thereby ensure high quality, easier-to-match photos in the database of authorized users. This quality thresholding will reduce both false positive and false negative errors in operation by enhancing the quality of probes and database images.

In addition to the ICAO standard, the ISO/IEC 29794-5 Face Image Quality standard is being updated and will become part of a new evaluation report provided by NIST FRVT.

To support these growing requirements for automated face image compliance checks, the ROC SDK now includes a complete automated ICAO compliance check. Specifically, the following factors can be measured and validated by ROC SDK algorithms:

FACTOR DESCRIPTION
Lighting Portraits shall have adequate and uniform illumination. Lighting shall be equally distributed on the face
Dynamic range The dynamic range of the image should have at least 50% of intensity variation in the facial region of the image.
Pose Head aligned toward the camera (Pitch/Yaw/Roll)
Expression The face shall have a neutral expression
Accessories: Glasses Tinted glasses, sunglasses, and glasses with polarization filters shall not be worn.
Accessories: Head coverings The region of the face, from the crown to the base of the chin, and from ear-to-ear, shall be clearly visible.
Portrait Dimensions + Head Location The head shall be centered in the final portrait
Portrait Dimensions and Head Location The image width A to image height B aspect ratio should be between 74% and 80%
Children Detect child age
Contrast For each patch of skin on the person’s face, the gradations in textures shall be clearly visible
Background Detect whether or not the background is uniform
Pose Eyes aligned toward the camera
Eye visibility Both eyes shall be opened naturally, but not forced wide-opened.
Accessories: Glasses Any lighting artifacts [e.g. glare] present on the region of the glasses shall not obscure eye details
Accessories: Facial Ornamentation Facial ornamentation which obscures the face shall not be present
Style: Makeup, Hair Style The hair of the subject shall not cover any part of the eyes
Rank One will continue to add to this range of compliance checks as new standards emerge.

Further, while these compliance checks can be integrated into existing or new applications via the ROC SDK, the ROC LiveScan application also provides a turn-key solution to performing image compliance checks and extracting the highest quality facial image of a person presenting themselves to a camera. This application is fully customizable to allow for additional ROC analytics to be included as part of the compliance check.

Video Analytics
Advancements in computer vision and machine learning technology have enabled a wide range of new possibilities in the automated analysis of images and videos. One of the most important emerging applications is the use of video analytics to support smart city initiatives, facility security, public safety, and traffic control.

With ROC 2.0 a wide range of new capabilities for video analytics are being delivered. These include object detectors, license plate recognition (LPR), and optical character recognition (OCR) capabilities.

Object Detection
The object detectors deployed with ROC 2.0 enable detecting the following objects:

  • Car
  • Truck
  • Bus
  • Motorcycle
  • Bicycle
  • Person
  • License Plate
  • Gun
  • Military Vehicle
  • Airplane
  • Boat

These detectors offer ROC integrators significant accuracy and efficiency improvements over other off-the-shelf solutions. While there is not an equivalent to FRVT for object detection, there are other methods to establish the effectiveness of the ROC object detectors. For example, when compared to the Yolo v5, a leading open source object detection framework in terms of both accuracy and efficiency, the new ROC detectors offer substantial performance improvements:

The detection rates presented above were measured using 1,000 images for each object detector from the OpenImages dataset. In addition to significantly accuracy improvements as compared to a leading open source solution, the ROC offering is also orders of magnitude faster:
With the ROC algorithm performing over 100x faster than leading open source alternatives while also delivering significant accuracy improvements, it is increasingly clear that ROC’s trade secret methods for delivering top-tier accuracy and efficiency will not just be limited to face recognition and will also set the industry standard for object detection and recognition.
Optical Character Recognition
In addition to object detection algorithms, the ROC SDK now includes the ability to perform Optical Character Recognition (OCR) across a wide range of challenging imagery. The solution is being used to power license plate readers, document processing, and a wide range of OCR-in-the-wild problems.

Stay tuned for more significant algorithm improvements ahead as Rank One Computing has officially entered the ROC 2.0 phase!

The post ROC SDK Version 2.0: A New Generation of Visual Analytics Algorithms  appeared first on ROC.

]]>
On Veteran’s Day, Rank One Computing (ROC) advocates to strengthen the Buy American Act to incorporate sensitive IT systems https://roc.ai/2021/11/10/on-veterans-day-rank-one-computing-roc-advocates-to-strengthen-the-buy-american-act-to-incorporate-sensitive-it-systems/ Wed, 10 Nov 2021 21:49:03 +0000 https://roc.ai/?p=5723 Honoring our Veterans who protect our Nation’s security, ROC is […]

The post On Veteran’s Day, Rank One Computing (ROC) advocates to strengthen the Buy American Act to incorporate sensitive IT systems appeared first on ROC.

]]>

Honoring our Veterans who protect our Nation’s security, ROC is raising awareness about the importance of securing the overall supply chain for our nation’s most sensitive IT systems. As the most trusted provider of facial recognition algorithms to U.S. military, law enforcement, and commercial organizations, ROC points to the fact that the U.S. is quickly developing trusted, domestic suppliers of Machine Learning algorithms. This comes as the U.S. is facing increased foreign attacks that exploit many of the legacy, foreign-developed algorithms. To mitigate these threats, the U.S. needs to support domestic suppliers of software that has “American nascency.”

 

Earlier this year, President Biden issued Executive Order, “Ensuring the Future is Made in All of America by All of America’s Workers” that questions whether the current economic and national security environment calls for the end to the 15-year exemption to the Buy American Act for commercial IT products. Removing this exemption would send a clear message that our national security infrastructure is taking these foreign threats seriously and is committed to developing our domestic AI/ML capabilities.

 

“Due to the gravity of potential harm that could arise from Artificial Intelligence-based attacks by foreign adversaries, trusted U.S. technology providers should be strongly preferred by federal government customers and domestic companies who take IT security risks seriously,” according to ROC General Counsel and Chief Operating Officer David Ray. “We frame the situation as follows and invite anyone interested in joining this conversation, to contact us.”

 

  • Technology in today’s world is taking on a strategic focus due to IT security risks posed by foreign adversaries.
  • Machine Learning algorithms are particularly prone to risks from “Poison AI” models that intentionally introduce untraceable security vulnerabilities into critical government systems.
  • Billions of dollars and unrestricted access to data are being provided by foreign adversaries to foreign companies with a goal of winning the technology race and thereby strategically positioning security vulnerabilities at the heart of the American economy and federal government.
  • To the extent that the exemption to the Buy American Act for commercial information technology was justified when introduced 15 years ago, that justification is no longer appropriate to today’s national interests and global technology landscape.
  • The exemption should be removed at minimum with respect to software such as facial recognition that powers critical government applications, and federal procurement should emphasize a strong bias in favor of using trusted, U.S.-made software solutions.

Press Release 

The post On Veteran’s Day, Rank One Computing (ROC) advocates to strengthen the Buy American Act to incorporate sensitive IT systems appeared first on ROC.

]]>