DEV Community: 3DiVi Inc.

Makeup as a Biometric Threat? How the “Living Vampire” Johnny Depp Fooled Liveness Detection

3DiVi Inc. — Mon, 21 Jul 2025 11:15:34 +0000

Facial recognition, especially with liveness checks, is often viewed as the ultimate defense in banking and fintech—able to stop everything from printed photos to deepfakes and 3D masks.

But what if the real vulnerability isn’t as high-tech as we think?

What if it comes in the form of… makeup?

In a recent internal experiment, the 3DiVi team uncovered a surprising blind spot—disguised in professional makeup.

We ran a video of Johnny Depp playing vampire Barnabas Collins in Dark Shadows through a liveness detection check. The result? The system confidently identified him as alive. And that’s a problem.

Why Makeup Is a Real Threat (Not Just a Theatrical One)

While most discussions around biometric spoofing focus on printed photos, video replays, or silicone masks, makeup rarely gets the attention it deserves.

Yet professional or theatrical makeup can:

Conceal or alter key facial features like eye shape, jawline, eyebrows, and mouth contours
Change the perceived skin tone and texture
Interfere with algorithms that rely on depth, color gradients, or texture cues

Unlike classic spoofing tools, makeup isn’t easy to classify. Where’s the line between everyday makeup and a deliberate spoofing attempt?

And more importantly, which appearance changes can biometric systems safely ignore — and which could open the door to fraud?

The Johnny Depp Test: How the Vampire Fooled the System

In our test, the video clip featured Johnny Depp’s character in full costume and makeup:

Extremely pale, chalk-like skin
Deep shadows around the eyes
Sharp, heavily defined facial contours
Dark lips and elongated brows

Despite the theatrical look, the system almost successfully recognized the face—and more importantly, it flagged the video as a live, real person. In other words, the liveness detection “brought the vampire to life.”

The clip wasn’t even live—it was just a pre-recorded video.

Is Makeup Now a Biometric Attack Vector

Our experiment proved just how easily a simple video with heavy makeup can bypass poorly configured or basic liveness detection systems.

Algorithms that depend solely on basic texture or motion cues are vulnerable to clever visual deception.

While most companies focus on defending against photo and replay attacks, cosplay, professional makeup, and even everyday cosmetics are turning into real tools for biometric spoofing.

To treat makeup as a biometric attack vector, certain conditions need to be met:

Repeatability: The failure should occur across different tests and subjects
Measured risk: The Attack Presentation Classification Error Rate (APCER) for makeup scenarios should be calculated
Thresholds: If APCER exceeds 1–2%, it’s already a red flag
Statistical significance: If spoofing with makeup results in significantly higher error rates (e.g., p < 0.05), it should be included in mandatory anti-spoofing tests

How to Test for Makeup-Based Spoofing

To make your biometric system resilient, consider:

Testing with a range of makeup: from light cosmetic to full theatrical looks
Including both images and video replays featuring makeup
Evaluating how your system responds to such inputs played from screens or mobile devices

Fraudsters don’t need expensive 3D masks anymore. Well-done makeup, paired with a screen and a video, is a low-cost but effective way to trick underprotected systems. If your liveness detection can’t handle this, it’s leaving a security gap open.

Closing the Makeup Vulnerability: Key Considerations for Biometric Teams

To avoid such vulnerabilities, companies should:

Demand that liveness detection accounts for depth, skin texture, and behavioral cues
Require proof from vendors that their system has been tested against spoofing scenarios involving makeup
Include spoofing resistance tests in their procurement and compliance checklists
Continuously update biometric security policies as new evasion techniques emerge

Final Thoughts: It's Not Just a Funny Glitch

The Johnny Depp vampire case isn’t a one-off bug or a funny story—it’s a signal. As biometric systems become more mainstream, they also become more attractive targets. And you don’t need Hollywood-level deepfakes to fool weak liveness detection. Sometimes, all it takes is a phone and some makeup.

To ensure biometric systems offer real protection, we need to test them under real-world conditions, including those that aren’t always obvious. In biometrics, just like in life, you have to know when reality ends and deception begins.

If you’re serious about closing hidden biometric security gaps, let’s talk. With 14+ years in computer vision and biometric security, 3DiVi knows how to spot and fix vulnerabilities before fraudsters find them.

Face Recognition in Flutter: Key Integration Insights You Need to Know

3DiVi Inc. — Fri, 18 Jul 2025 07:55:49 +0000

Face recognition technology is rapidly becoming a common part of mobile app experiences, enabling faster digital onboarding, accurate authentication, and personalized interactions that improve UX and strengthen customer data security.

In the U.S. alone, around 132 million people use face recognition on at least one app daily. The adoption rate is especially high among younger generations, with 75% of 18- to 34-year-olds incorporating face recognition into their daily lives, and 57% of them using it every day.

While 68% of usage still comes from unlocking phones and tablets, face recognition is quickly becoming the preferred login method for banking, healthcare, ticketing, and other security-sensitive apps — with 51% of users relying on it to access these services.

But while demand is rising, implementation comes with its challenges. Integrating face recognition into a mobile app depends heavily on your tech stack — and for developers working with Flutter, that raises specific considerations.

In this article, we break down exactly how to add face recognition into your Flutter app — step-by-step. Whether you’re evaluating SDKs, worried about performance, or just want to future-proof your app, we’ll walk you through everything you need to know to get a successful integration.

What is Face Recognition and Why Use It in Your Flutter App

Face recognition is a form of biometric authentication that uses computer vision and machine learning to identify or verify a person based on facial features typically taken from a photo or a camera feed.

The process involves detecting a face, identifying key anthropometric points, generating a biometric template, and comparing it against already stored templates to find a match.

As mobile security becomes increasingly critical, integrating face recognition into apps is no longer a luxury — it’s a necessity.

Flutter, with its cross-platform capabilities and unified codebase approach, makes this integration efficient and scalable, enabling developers to implement face recognition pipeline once and then deploy it effectively across both Android and iOS devices.

Here’s what makes face recognition in Flutter a smart move:

1. It’s What Users Expect

Face unlock has become second nature on smartphones. Whether it’s logging into a banking app or confirming a payment, users now expect that same instant, touch-free experience in mobile apps. Integrating face recognition into your Flutter app meets this expectation head-on—and gives your product a competitive edge.

2. Security Without the Hassle

Face recognition helps prevent fraud, unauthorized access, and identity spoofing (especially when combined with liveness detection). For fintech, healthtech, or any app handling sensitive data, this adds a frictionless security layer that doesn’t compromise UX.

3. Cross-Platform Support, One Codebase

Flutter allows you to build for both Android and iOS from a single codebase. When paired with a cross-platform face recognition SDK or native integration strategy, you get consistent biometric functionality across devices—no need to write separate native modules.

How to Integrate Face Recognition into Your Flutter App

Choosing the Right Face Recognition Software

Before diving into coding, the first step is to choose a reliable face recognition SDK (Software Development Kit) or API that supports Flutter. There are two main types of plugins available:

Open-Source Plugins

You can find many open-source face recognition plugins on pub.dev, the official package repository for Dart and Flutter developers. Pub.dev allows developers to discover, share, and manage open-source packages, libraries, tools, and plugins to extend the functionality of their projects.

However, finding a reliable face recognition solution on pub.dev can be challenging. While a quick search for "Face Recognition" might return over 100 plugins, not all of them meet your specific needs.

To help you choose the best one, check out our article, "Top Face Recognition Plugins in Flutter: What's Actually Worth Your Time on pub.dev?".

Commercial Plugins

These plugins offer ready-to-use solutions with robust support and a broader range of features than many open-source options. They are often a better choice if you need more advanced capabilities and professional support.

Once you've selected the appropriate face recognition SDK with Flutter support, you can move on to the integration process.

3 Integration Bottlenecks and How to Avoid Them

Integrating face recognition into Flutter apps isn't always straightforward. Here are a few real-world challenges we've encountered — and how we solve them.

Immutable Objects and Image Conversion Performance

Object immutability is a core concept for Flutter’s core language, Dart, meaning developers typically can't modify objects without creating a copy.

While this has benefits for stability and predictability, it can seriously impact performance when working with images—especially during conversions from camera formats to RGB.

To address this, we often rely on native (C++) image conversion implemented directly within our Face SDK when working in Flutter.

This approach significantly optimizes image processing operations. In particular, using the C++ version of image conversion has helped us boost FPS several times over.

Using Dart Isolates for Heavy Processing

Heavy processing — including Face SDK inference — can often interfere with app performance, especially the UI.

This can lead to lags, stutters, or even app freezes in more extreme cases. To avoid these issues, we recommend using Dart Isolates to offload demanding tasks to separate threads.

To make this easier for developers, we’ve already integrated Isolate support directly into our Flutter plugin.

Here’s a less obvious but equally important scenario:

In some apps, you may need to save cropped face images — for example, to display them as thumbnails in a database of registered users. However, converting an RGB image to JPEG can take up to 200–250 ms.

If you attempt this without using Isolates, the camera preview may freeze or stop working entirely. With Isolates, the app stays responsive.

Handling YUV_420_888 to RGB Conversion on Android

When working with android.hardware.camera2 — the standard package for camera interaction on Android — the image format YUV_420_888 is commonly used. However, for neural network processing, we often need to convert these images to RGB.

As mentioned earlier, we handle this conversion in native code using our custom implementation. This generally works well — but on some devices (such as the Honor X8b), the image data doesn’t fully comply with the expected format.

As a result, the converted RGB image becomes corrupted, and the processing pipeline fails.

That’s why we continuously maintain and update our Flutter Face SDK to ensure compatibility across a wide range of smartphones.

The best part?

All of these edge cases and performance hurdles are already solved in 3DiVi Face SDK for Flutter—built to accelerate face recognition integration and keep your app running smoothly on real-world devices.

Why Choose 3DiVi Face SDK for Flutter Apps?

Easy Integration for Flutter Developers

With 3DiVi Face SDK, developers can integrate face recognition functionality without needing to be computer vision experts. The SDK is designed with ease of use in mind, allowing developers to add face recognition features in just a few lines of code. Thanks to comprehensive documentation, including step-by-step tutorials and samples, even beginners can get started quickly.

Flexible Face Recognition Pipeline with Additional Options

3DiVi Face SDK for Flutter offers a versatile face recognition pipeline, capable of handling a variety of tasks, including:

Face Detection: Accurately detect faces in real-time, even in challenging conditions like low light or occlusions.
Age, Gender, and Emotion Estimation: Estimate demographic details and detect emotions, adding depth to user interactions and analytics.
Liveness Detection: Strengthen security with liveness checks to confirm the user is a real person, not a spoofed image or video.
Image Quality Assessment: Evaluate facial image quality to ensure optimal face recognition performance.
Face Identification & Verification: Perform fast and accurate 1:N and 1:1 face matching.

Efficient Use of Resources with Isolates

The SDK provides Dart Isolates functionality to offload compute-intensive tasks, ensuring smoother app performance. By executing biometric operations on separate threads, it eliminates UI freezing and lags and improves responsiveness during face recognition processing. This brings app stability, even on budget devices, providing a frictionless experience for identity verification solutions.

Constant Updates and Support

With regular updates and continuous improvements, we ensure that 3DiVi Face SDK stays at the forefront of face recognition technology. Access up-to-date tutorials, detailed documentation, and technical support to keep your integration running smoothly.

Thinking of integrating face recognition into your Flutter app? Get started with 3DiVi Face SDK for Flutter by following our detailed tutorials and samples. Schedule a free consultation to explore how fast, accurate, and secure face recognition can improve your app!

Two Face Recognition Projects Failed. $33K Burned — All Because of Bad Camera Setup

3DiVi Inc. — Mon, 07 Jul 2025 07:15:44 +0000

In video analytics scenarios like Safe Cities, Access Control, or Retail facial recognition software only performs at its best when the cameras are set up correctly.

You can have world-class facial recognition algorithms, premium servers, and top-tier cameras — but one misaligned lens or wrong exposure setting can bring the entire facial recognition system down.

Here are two true stories where poor camera setup didn’t just cause hiccups — it completely derailed multi-thousand-dollar projects.

❌ Story #1: “Smart” Traffic Lights That Never Worked

A city project set out to launch smart traffic lights equipped with built-in cameras to monitor situations and recognize faces on city streets.

But despite the ambition, the execution fell short: the cameras were installed and configured “by eye”, without proper planning, calibration, or performance testing for AI facial analysis.

Unsurprisingly, the facial recognition system performed poorly during acceptance tests — especially in the evening, when backlighting and bad weather significantly degraded image quality and face identification accuracy.

A thorough review of interim system data could have exposed the flaws early on. But under heavy workload, the contractors didn’t have the time for in-depth analysis.

The result? A 9-month pilot from June 2023 to March 2024, 15,500 US dollars spent — and no viable product. The project was declared unsuccessful, and the team failed to enter the market with a working smart traffic light solution.

❌ Story #2: Face Identification in a Mall = Lost Contract + Fired PM

An facial recognition system was set to launch in a major shopping and entertainment center. The project dragged on for over a year and cost 17,600 US dollars.

The cameras were installed according to plan, but the settings — exposure, noise reduction, resolution — were adjusted “by eye” by the project manager. No full-scale system checks. No performance validation. Just a quick confirmation that some AI face matches were coming through — and the team moved on.

But on testing day, reality hit. The client ran 50 control walkthroughs. Face identification failed to perform — and so did the project. The contract was canceled, and the project manager was suspended.

These projects didn’t flop due to bad facial recognition algorithms or weak hardware — they failed just because camera placement and settings weren’t properly validated for facial recognition tasks.

We’ve seen this story too many times.

In our hands-on work with both outdoor and indoor video analytics projects, even small camera mistakes could lead to massive recognition failures.

That’s why we dug deeper and identified 14 critical factors that directly impact performance of facial recognition technology:

1. Network Bandwidth
2. Hardware Stability
3. Camera Resolution
4. Camera Sensor Quality
5. Video Analytics Server Performance
6. Quality of Reference Photos
7. Vibration (Wind, Vehicle Movement)
8. Weather Conditions (Rain, Fog, Snow)
9. Distance to Target
10. Crowd Density
11. Speed of Movement
12. Backlighting (Sunlight, Reflections)
13. Camera Angle and Position
14. Illumination Level (<200 lux)

Read the full breakdown in our article: 14 Factors You Can Control to Improve Face Recognition Efficiency in Safe Cities.

Trying to account for all of these manually? Practically impossible. It takes expert-level training, meticulous tuning, and a ton of time — something most integrators don’t have.

So we built 3DiVi Cam QA — an automated camera setup assessment tool that analyzes live or archived footage and generates a report across 19 key parameters, with clear recommendations to optimize camera placement and configuration for the best facial recognition performance.

Today, 3DiVi Cam QA is already helping integrators protect their deals, clients, and reputation. Want to see how it works? Let’s connect.

Goodhart’s Law in AI: How to Avoid the Metrics Trap in Facial Recognition Projects

3DiVi Inc. — Thu, 19 Jun 2025 10:55:17 +0000

"When a measure becomes a target, it ceases to be a good measure."

That’s Goodhart’s Law, introduced by British economist Charles Goodhart in 1975 — and it’s more relevant today than ever, especially in the age of AI.

AI systems, when optimized solely for specific performance metrics, often end up serving the metric instead of the real goal.

Let’s break down how this plays out in real-world AI applications — and how we avoid this trap in our AI facial recognition technology.

AI in Education: Teaching to the Test

Imagine an AI-powered tutoring system evaluated by how many correct answers students get on tests.

Sounds logical — until you realize the system might begin prioritizing rote memorization over actual learning.

The result? Students may ace the tests but lack critical thinking or creative problem-solving skills. AI meets its metric, but misses the point of education.

AI in Healthcare: More Procedures ≠ Better Outcomes

Now take healthcare. If diagnostic AI is judged by the number of tests or surgeries it leads to, it might start recommending unnecessary procedures just to hit the numbers.

This not only wastes resources—it can actively harm patients. The metric is satisfied, but at what cost?

AI in Business: The Sales Trap

AI is frequently used to boost sales. But when its performance is measured purely by transaction volume, the system might push deals that aren’t sustainable—offering steep discounts, or focusing on leads unlikely to convert long term.

It might spike short-term revenue, but erode profitability and customer trust in the long run.

AI in Law Enforcement: Misplaced Focus

Some law enforcement agencies use AI to predict where and when crimes might occur. If success is defined by the number of predicted crimes, the algorithm might start flagging minor infractions—just to meet its quota.

This leads to over-policing in low-risk areas, while real threats go unnoticed. Again, the metric is gamed, not the mission achieved.

How Do We Avoid Goodhart’s Trap in Our AI Face Recognition Projects?

🔹 We evaluate AI face recognition models using a wide set of KPIs, including robust industry standards like those from NIST. No single number tells the whole story.

🔹 We test models in the wild, not just on "clean" datasets. Real-world scenarios—bad lighting, occluded faces, network instability—are where real performance matters.

🔹 We continuously re-evaluate goals and confidence thresholds.
What counts as “good” depends on the use case: AI facial recognition software for access control, a banking app, or a transit system all need different thresholds. We adapt based on feedback from integrators and end users.

Final Thought: Metrics Aren’t Bad—But They Can Backfire

Goodhart’s Law is a powerful reminder: if your AI is chasing numbers, it might stop solving real problems.

To make AI work in real-world applications, we need to build and evaluate systems that align with outcomes, not just indicators.

9 Key Criteria for Selecting The Best Facial Recognition Software in 2025

3DiVi Inc. — Thu, 05 Jun 2025 12:06:31 +0000

Choosing the right AI facial recognition software is mission-critical for businesses aiming to integrate biometric technology into their products or systems in 2025.

Whether you're in banking & fintech, security, healthcare, or retail, making the wrong decision can lead to poor accuracy, privacy violations, and costly integration failures.

This guide outlines 9 essential criteria B2B buyers must evaluate to make the right software selection.

1. Accuracy

High accuracy of the AI face recognition system ensures that legitimate users are correctly recognized while keeping impostors out — reducing fraud, user frustration, and onboarding failures.

To decide whether two faces match, AI facial recognition technology relies on a similarity score threshold that balances two types of errors:

False Acceptance Rate (FAR): The percentage of unauthorized faces incorrectly accepted as legitimate. Also known as False Match Rate (FMR) for 1:1 verification or False Positive Identification Rate (FPIR) for 1:N identification (NIST FRTE terms). A high FAR increases the risk of security breaches and compliance violations.
False Rejection Rate (FRR): The percentage of legitimate faces incorrectly rejected. Also called False Negative Match Rate (FNMR) for 1:1 or False Negative Identification Rate (FNIR) for 1:N (NIST FRTE terms). A high FRR causes user frustration, increased drop-offs, and failed onboarding.

This trade-off is typically visualized as a curve showing how adjusting the similarity threshold affects both error rates:

Tight thresholds lower FAR (fewer impostors accepted) but increase FRR (more real users rejected).
Loose thresholds lower FRR (smoother user experience) but raise FAR (greater risk of unauthorized access).

3DiVi’s AI facial recognition technology (12v1000) uses a recommended similarity threshold of 0.85, optimized for both security and usability:

FAR: 0.0000009919 — nearly zero chance of unauthorized entry
FRR: 0.0075107813 — minimal disruption for real users

This makes it a strong fit for most real-world deployments, from digital onboarding to access control, where both trust and ease of use are critical.

✅ Tip: Choose vendors who provide transparent accuracy benchmark results on reputable datasets like NIST FRVT. These offer a real-world measure of how their algorithm performs (e.g., see 3DiVi facial recognition software accuracy benchmarks).

2. Processing Speed

Whether it's unlocking a mobile app, approving a face payment, or opening a secure door, delays — even brief ones — can frustrate users, leading to entry-point queues, stalled workflows, or abandoned checkouts. That’s why top facial recognition software need to be fast at every stage:

Detection Speed

This is how quickly the AI face recognition system can locate a face in a photo or video stream. Faster detection allows for smoother, more responsive interactions — crucial for applications like live surveillance, mobile face unlock, or turnstile access in real time.

Template Generation Speed

After detection, the system must extract unique facial features and convert them into a face biometric template. This process should be fast enough to avoid delays during user enrollment or onboarding.

Verification Speed (1:1 Matching)

In 1:1 scenarios, the AI face recognition system compares the user’s current facial template to a stored one. Fast face verification directly impacts user experience by reducing wait times — especially in mobile or access control use cases where instant response is expected.

Identification Speed (1:N)

In identification scenarios, the AI face recognition system compares the detected face against a database of face biometric templates to determine identity. This is common in law enforcement, public safety, or enterprise-grade access control, where timely identification from many records is required.

✅ Tip: When comparing vendors, ask for detailed performance benchmarks covering detection speed, template generation, and face match time. These metrics should align with your specific needs — whether you require post-event video analysis or instant authentication on mobile devices (e.g., explore 3DiVi facial recognition software speed test results).

3. Integration Ease

When it comes to AI facial recognition software, how easily it fits into your existing systems can make all the difference. Smooth integration accelerates development, reduces costs, and ensures faster time-to-market.

APIs and SDKs: Your Building Blocks

Opt for vendors that offer well-documented, easy-to-use face recognition APIs and / or SDKs, which provide the building blocks for embedding AI facial recognition technology into your applications. The ease of use, clarity of documentation, and robustness of these tools directly impact developer productivity and integration speed.

Supported Platforms

Ensure the AI facial recognition software supports the major operating systems your business relies on, including Windows, Linux, Android, or iOS. Broad platform compatibility enables deployment across diverse devices, from servers to mobile endpoints.

Supported Programming Interfaces

Check that the vendor provides face recognition SDKs or libraries compatible with popular programming languages such as C++, C#, Java, or Python. This flexibility allows your development team to work within familiar environments, simplifying integration and maintenance.

✅ Tip: Choose vendors that deliver comprehensive, user-friendly integration tools with broad platform and language support — so you can add AI facial recognition software to your technology stack without disruption.

For example, 3DiVi facial recognition software is designed for cross-platform compatibility, supporting Windows, Linux, Android and iOS — and a broad spectrum of programming languages, including Python, C++, C#, Kotlin, Flutter, Swift, and Java.

4. Security Features

Facial biometric data is highly sensitive and tightly regulated, so your software must handle it with the highest level of care.

Data Privacy Compliance

Make sure your chosen solution fully complies with key regulations like GDPR (General Data Protection Regulation), CCPA (California Consumer Privacy Act), or AML (Anti-Money Laundering). Compliance means your users’ biometric data is processed legally and ethically, safeguarding privacy and minimizing legal risks.

Certifications and Audits

Trust vendors who hold recognized security certifications—such as ISO/IEC 27001 or SOC 2—or undergo independent third-party audits. These credentials prove the company meets rigorous security standards, giving you confidence that your data and systems are protected.

✅ Tip: Pick the best facial recognition software that combine legal compliance with industry-standard security certifications — so your biometric data stays secure and your organization stays on the right side of the law.

For instance, 3DiVi facial recognition software is fully compliant with GDPR, CCPA, KYC, and AML regulations.

5. Scalability

As your user base and data grow, top facial recognition software must keep up—scaling without sacrificing speed or accuracy.

Database Size and Capacity

Check how well the face recognition solution handles large face databases. A strong system should manage millions of identities without slowing down or losing reliability.

Processing Load Handling

Evaluate whether the AI face recognition system can process recognition requests simultaneously, especially during peak times. Maintaining consistent performance under heavy load is vital for critical applications.

Auto-scaling

Look for systems that automatically adjust resources based on demand. Auto-scaling helps balance performance and costs by expanding or shrinking infrastructure as needed.

Integration with Cloud Services

Support for major cloud platforms like AWS, Microsoft Azure, or Google Cloud allows flexible, scalable deployments. Cloud compatibility simplifies management and ensures global accessibility.

✅ Tip: Choose face recognition solutions that scale with your business—whether you're managing millions of identities or handling spikes in authentication traffic. Look for platforms that support cloud deployment, auto-scaling, and reliable performance under heavy loads.

For example, 3DiVi facial recognition software supports deployment on AWS and Google Cloud, with proven performance under peak loads.

6. Cost

Don’t let hidden costs derail your project. Evaluate the total cost of ownership upfront to ensure the facial recognition integration stays within budget—both at launch and as it scales.

Subscription Fees

Many vendors offer subscription-based pricing tied to usage levels. Evaluate how these ongoing fees scale with your usage patterns.

Implementation and Integration Costs

Consider expenses related to deploying the software and integrating it with your existing systems, including developer time and customization efforts.

Maintenance and Support Fees

Ongoing costs for technical support, software updates, and bug fixes can significantly impact your budget over the product life cycle.

Scalability Costs

Scaling up the system to handle more users or data often incurs additional charges. Plan for these costs as your deployment grows.

Operational Costs

Don’t overlook recurring expenses such as cloud storage, data transfer fees, and infrastructure costs that support the software operation.

Hidden or Additional Costs

Watch out for extra fees for advanced features, premium support, or unforeseen charges that might arise after deployment.

✅ Tip: Choose transparent pricing models that align with your business needs and provide clear visibility into all potential costs to avoid budget overruns.

7. Support and Maintenance

Reliable support and ongoing maintenance are a must to ensure smooth operation and quick resolution of issues in face recognition deployments.

Support Availability

Evaluate the vendor’s support hours and ensure they align with your operational schedule, especially if your system requires 24/7 uptime. Rate vendors based on how well their support availability matches your needs.

Response Time

When something breaks, how fast can you expect help? Check service level agreements (SLAs) and customer reviews for insights into real-world response and resolution times.

Documentation and Resources

Vendors that provide clear manuals, developer guides, FAQs, and integration tutorials empower your team to troubleshoot independently—speeding up onboarding and reducing reliance on external support.

✅ Tip: Select top facial recognition software from vendors who offer timely, accessible support backed by rich documentation to reduce operational risks and improve user experience. For example, all 3DiVi products have detailed technical documentation at https://docs.3divi.ai/

8. Deployment Flexibility

Deploying facial recognition software in a way that fits your infrastructure and operational needs is one more major factor in vendor selection.

Cloud vs. On-Premises

Choose the software that supports your preferred environment—whether that’s cloud for quick scalability and remote access, on-premises for full control and data security, or a hybrid model that blends the best of both.

Disaster Recovery and High Availability

Look for features like automated backups, failover support, and system redundancy. These ensure your face recognition solution stays resilient—even during outages.

✅ Tip: Prioritize vendors that offer flexible deployment models, setup, and strong disaster recovery features — so your system is both secure and scalable from day one.

9. Customizability

Off-the-shelf solutions rarely fit every use case—customization is what transforms them into a precise match for your needs.

Customization Support

Check whether the vendor allows modifications to features, workflows, or user interfaces to better fit your unique use cases and operational requirements.

Additional Costs

Be aware of any extra fees associated with customization or bespoke development services. Transparent pricing helps avoid unexpected expenses.

✅ Tip: Choose vendors that provide flexible, customizable solutions capable of adapting to your specific business needs—without hidden costs or long delays.

Best facial recognition software is not a one-size-fits-all solution. With evolving compliance standards, diverse deployment environments, and performance expectations, selecting the right technology partner is a strategic decision—not just a technical one.

For over 14 years, 3DiVi Inc. has been helping businesses worldwide integrate AI-powered facial recognition across various industries—from security and fintech to retail and public safety. Let's discuss how our facial biometric solutions can drive measurable impact for your business.

How to Port CV/ML Models to Rockchip NPU for Faster Face Recognition

3DiVi Inc. — Tue, 03 Jun 2025 06:14:10 +0000

In 2024 the 3DiVi team faced a new challenge: one of our partners decided to build an access control system (ACS) on a single-board computer from Forlinx.

To meet the strict time constraints for face recognition, we ported our models to the NPU. Long story short—it worked! NPUs turned out to be a solid way to put heavy processing to an edge device.

How Face Recognition Works: A Quick Recap

Skipping the business details, here’s what our partner needed: detecting faces in a video stream and verifying them. To better understand what that means, let’s take a closer look at the basic face recognition pipeline.

Face Detection:

The detection module identifies a face in an image. Most face detectors in production today rely on convolutional neural networks (CNNs).

Key Point Detection:

Next, key points (like the eyes and nose) are located on the face. This is also done by a neural network—sometimes a separate one (called a "Face Fitter") or the same one used for detection. In our case, we use a dedicated CNN.

Face Alignment:

Using the detected key points, the face is aligned to a frontal position, a necessary step in biometric template generation.

Template Extraction:

Finally, another neural network extracts the biometric template from the aligned face crop.

Speed and Hardware Constraints

In most scenarios, we can overlook the time spent on image preprocessing, postprocessing of neural network results, and comparing two biometric templates—these are just a few milliseconds. The bottleneck? Neural network inference time.

In our case, there were three neural networks:

Face Detector
Face Fitter
Face Template Extractor

And we were working under these time constraints:

Combined time for Face Detector and Face Fitter: ≤40 ms.
Template extraction and comparison of two templates: ≤500 ms.

Sounds manageable, right? But then we looked at our hardware: OK3568-C. Not exactly ideal for heavy processing.

We selected specific models for the detector, fitter, and template extractor from 3DiVi Face SDK and tested their inference times. As expected, the results shown in the table below didn’t meet the stated time constraints.

After that, we moved on to inference on the NPU.

Rockchip NPU Inference

Rockchip NPU inference can run in two modes:

Default Mode: Models are converted from Float32 to Float16. This leads to minimal (often negligible) accuracy loss.
Quantized Mode: Models are converted from Float32 to Int8. This significantly speeds up inference but can result in noticeable accuracy drops.

During the experiments, we obtained the following time measurements:

But what about accuracy? Yes, there was a slight dip. However, after some fine-tuning, the Int8 quantized model performed well enough for production on a standard dataset (LFW).

Final Thoughts

Porting CV/ML models to an NPU proved to be an effective way to accelerate inference. The minor accuracy drop — acceptable for access control systems — was worth it to meet the recognition speed requirements.

6 Threat Actors Targeting Face Authentication in 2FA / MFA

3DiVi Inc. — Mon, 02 Jun 2025 10:20:54 +0000

From digital banking to e-commerce, face recognition is now a common layer in 2FA and MFA stacks. But as adoption rises, so does risk.

Behind every spoofed identity is a threat actor—and they’re more diverse than you think. Before we turn into the six types of threat actors exploiting face authentication vulnerabilities, let’s first explore why knowing this categorization is essential for strengthening your security posture.

Why This Matters for Business Cybersecurity

Prioritize Your Defenses: Not all threats are created equal. State-sponsored actors demand a different response than script kiddies or insider risks. Understanding who is most likely to target your system allows you to focus defenses where they matter most and avoid spreading resources too thin.
Pressure-Test Your Vendors: Are your face recognition providers regularly testing against the latest face biometric threats? Do they offer robust liveness detection and anomaly scoring? If not, you may be leaving your systems open to more than just theoretical risks.
Adjust Authentication Scenarios Based on Threats: By identifying the most likely threat actors targeting your system, you can change your authentication scenarios accordingly. For example, if you're dealing with criminal hacking syndicates, you might implement environmental controls + liveness PAD during high-value transactions, while adapting simpler flows for low-risk interactions (login attempts from known devices or password resets).
Future-Proof Compliance: Regulatory bodies are increasingly scrutinizing the use of biometrics in identity verification. Knowing the vectors of attack today prepares you to meet the security, privacy, and audit requirements of tomorrow.

6 Threat Actors Targeting Face Authentication in 2FA / MFA

1. Nation-State Actors

State-sponsored threat actors, operating on behalf of national governments, primarily engage in cyber activities to achieve their geopolitical objectives.

Whether it’s surveillance, destabilization, or long-term espionage, these players bring serious resources to the table: skilled personnel, custom-built tools, and the patience to spend months—or even years—on a single campaign.

National states without advanced cyber programs often outsource to contractors, buy access to commercial hacking tools, or partner with organized criminal groups to get the job done.

2. Cyber (Digital) Mercenaries

Cyber mercenaries operate for whoever pays them the most: governments, corporations, or even shady third parties.

Their job? Anything from stealing trade secrets and launching espionage campaigns to knocking out infrastructure remotely across different jurisdictions, including both defensive and offensive cybersecurity operations.

Think of them as freelancers for digital warfare—highly skilled, highly motivated, and not bound by borders or ethics.

3. Criminal Hacking Syndicates

Criminal hacking syndicates work like digital mafias. They run phishing campaigns, spread malware, buy and sell breached data, and coordinate large-scale fraud.

But unlike lone hackers, these groups are global, collaborative, and built for scale.

Organized cybercrime continues to evolve and adapt as these syndicates develop increasingly sophisticated methods for exploiting sensitive personal and business data.

4. Technical Stalkers

Not all cyberattacks are about money or ideology—some are deeply personal.

Technical stalkers are individuals who use hacking skills to pursue private agendas: revenge, obsession, or harassment. This group often includes disgruntled former employees, rejected partners, or individuals with personal vendettas.

What sets them apart is not just intent, but persistence. Unlike opportunistic attackers, they may invest significant time and effort in their target—using advanced biometric spoofing techniques to compromise a specific person or organization. Their attacks don’t always make headlines, but they can be deeply damaging.

5. Script Kiddies

The term refers to inexperienced individuals who use pre-made tools, scripts, or tutorials to launch cyberattacks. They typically lack deep knowledge of systems, networks, or security architecture. Instead of writing their own exploits, they rely on what others have built.

While their technical skills are limited, the threat they pose shouldn’t be underestimated. With access to readily available spoofing kits and emulator software, even a novice can attempt to bypass facial recognition or overload verification systems with repeated attacks.

6. Unintentional Legal Users

Legal users can sometimes gain access to sensitive information unintentionally, often due to design flaws or the probabilistic nature of face recognition systems. These incidents may occur without any malicious intent but still pose significant security risks.

Cybersecurity is no longer just about defending against generic attacks — it’s about knowing your adversaries and how they operate. From nation-state actors to unintentional insiders, each poses a different risk and calls for a specific strategy.

👉 Action Step: Review your current biometric authentication setup against these six threat profiles. Then align your vendor criteria, user flows, and compliance roadmap to defend against today’s real-world risks — not outdated assumptions.

Do We All Have a Lookalike? Face Recognition vs. Human Eye

3DiVi Inc. — Fri, 30 May 2025 12:08:07 +0000

Since 1999, Canadian photographer François Brunelle has been chasing a fascinating mystery: Can two complete strangers look like identical twins?

His ongoing photo project “I’m Not a Look-A-like!” features striking black-and-white portraits of unrelated people who resemble each other so much, they could easily pass as twins. Over the years, Brunelle has photographed more than 250 such pairs in 32 cities around the world.

But how does facial recognition technology see these “twins”? We decided to find out.

We ran several of Brunelle’s look-alike portraits through our online facial recognition demo.

The result? The algorithm didn’t find a single match—even in cases where most people would swear it was the same person.

Let’s take a closer look at how the experiment worked — starting with how a face recognition algorithm decides: “same person” or “no match”.

How Face Recognition Knows Who’s Who

To make a decision on whether two faces match, the algorithm relies on a trade-off graph between two types of errors:

False Rejection (FRR):
A real user is wrongly denied.
→ Leads to frustration, drop-offs, failed onboarding.
False Acceptance (FAR):
An imposter is wrongly accepted.
→ Leads to security breaches and compliance risks.

These are measured via a similarity score threshold. Tighten the threshold = fewer imposters get through, but more real users get blocked. Loosen it = smoother UX, but higher risk.

The default recommended similarity score threshold for our algorithms is 0.85. At this threshold, the algorithm achieves:

False Acceptance Rate (FAR): 0.0000009919
False Rejection Rate (FRR): 0.0075107813

This balance ensures extremely low chances of unauthorized access while maintaining high acceptance for legitimate users.

Now let’s move on to specific examples taken from the website http://www.francoisbrunelle.com/webn/e-project.html

What the Lookalike Analysis Revealed

Example 1.

Score is 17. FAR=0,019458; FRR=0,003017. Verdict: Different people.

Example 2

Score is 23. FAR=0,007506; FRR=0,003363. Verdict: Different people.

Example 3

Score is 14. FAR=0,053261; FRR=0,002691. Verdict: Different people.

And for a clearer visualization — here's a comparison of individuals taken from different lookalike pairs.

Score is 15. FAR=0,034590; FRR=0,002873. Verdict: Different people.

These results show that even when two faces appear strikingly similar, face recognition algorithms can still distinguish between them with high accuracy — far outperforming human judgment in such cases.

But here’s the catch: accuracy isn’t guaranteed. It still hinges on a few critical factors — like quality and diversity of training data, the underlying model architecture, and the real-world conditions.

What Really Impacts Face Recognition Accuracy

Despite impressive results, face recognition performance still depends on several key factors:

Training data quality: The more diverse and extensive the data used to train an algorithm, the better it performs in recognizing different types of faces (varying in age, race, and gender). High-quality and well-balanced datasets significantly boost accuracy.
Model architecture: Modern face recognition algorithms—especially those based on convolutional neural networks (CNNs)—achieve high accuracy thanks to deep learning and the ability to detect subtle facial features. Nearly all leading market players use complex neural architectures for precise facial identification.
Image quality: Just like with human recognition, image clarity is critical. Sharp images with good lighting significantly improve recognition accuracy, while blurred, dark, or partially obscured faces can challenge the system (Tip: Tools like 3DiVi’s QAA can pre-filter low-quality images before they hit your system).
Appearance changes: Contemporary algorithms are capable of recognizing faces despite minor appearance changes (e.g., hairstyle, makeup, or glasses). However, drastic alterations can make recognition more difficult.
Cross-race effects: Just like humans, algorithms may be biased depending on the data they were trained on. If the training set lacks ethnic diversity, algorithms may struggle to recognize underrepresented groups. Still, even accounting for this effect, modern systems make significantly fewer mistakes than humans in similar conditions.

Where to Find Performance Metrics

FRVT (Face Recognition Vendor Test): Tests on various datasets show that in 1:1 verification scenarios (e.g., matching a passport photo to a person), top systems have an error rate of less than 0.01%. In 1:N identification tasks (e.g., searching a face in a database), accuracy remains high but depends on database size and image quality.
Real-time face recognition: In challenging environments (e.g., outdoor surveillance), accuracy can drop. However, advanced algorithms still achieve over 95% accuracy, particularly when improved image processing and adaptive techniques are applied.

Examples of Real-World Systems

Apple Face ID delivers 98–99% accuracy in optimal lighting and angles — impressive for a consumer device.
3DiVi algorithms, benchmarked by NIST, show world-class performance with a False Match Rate (FMR) of 0.000001 and False Non-Match Rate (FNMR) of 0.003 at the default threshold.

Final Thoughts

Modern face recognition algorithms are pushing the boundaries—reaching near-100% accuracy and outperforming humans in controlled settings. But it’s not without limits — bias in training data, false match rates, and poor image quality can still get in the way.