Goals Enable consumers & enterprises to move to a secure, password free world – inclusive • of device unlock, payments, and secure content Develop hardware solutions that meet or exceed customer expectations for • seamlessness, robustness, and ease of use Deliver biometric devices that embrace innovation and deliver customer value • Create a more personal computing experience that makes technology interaction • natural, rather than intrusive, so users can quickly get to work, or play
:) Build Vector Detect head Does it match a Discover Find a Face based Template? Landmarks Orientation Representation Head orientation 1. Find face & discover landmarks 2. Representation vector 3. Decision engine 4.
Step 1: Head Orientation (Frontal Face) Step 2: Find Face & Discover Landmarks Step 3: Representation Vector Step 4: Decision Engine
T est environment/lab setup and testing process is outlined providing guidance on how to meet the current specification requirements
Frame MTF Filter Uniformity Gamma Ambient Saturation IRSNR MTF 50 MTF 50 Distortion Pairing Over/Undershoot Incandescent light face in illuminated > 30/26/22 Full .25 < cy/pxl < 5% / 3% Pixel to FOV @ far range .25 < cy/pxl 30nm+- Yes/No < 65% @ For ALS (Ambient frame can't be @ far range @ far range Reflectivity < 5.5% Mid-range Light Subtraction) saturated R² > 0.98 test @ near/far range Center Corner Center Center Corner Over Under 45% 0.99 31 30 0.256 0.254 0.25 3.2% 1.0% 2% 0lux No ALS Pass 27 26 0.223 0.23 0.21 4% 2% 50lux ALS Pass Yes pass 23 22 0.2 0.21 0.2 5% 3% 150lux ALS Pass 300lux No ALS Pass
T est Environment 2
Image Signal Processing T uning ISP tuning artifacts interfere with face authentication algorithms
100 aoi=0 90 aoi=30 80 70 Optical Transmission (%) 60 50 40 30 20 10 0 400 500 600 700 800 900 1000 1100 Wavelength (nm) Need to ensure that we have sufficient IR illumination at edge of sensor – impacts IR SNR, and overall device performance
Microsoft Pixel-by- Artifact Ambient Pixel Simple pixel-by-pixel ambient subtraction can Subtraction Subtraction yield image artifacts Light-field Artifacts Microsoft algorithm aligns marker positions to perform ambient subtraction Artificial Key Advantages: Shading Motion invariance • Reduction in artifacts Dark • Marking Requirement: 15 FPS for Ambient and Illuminated frames • Poor Edge Delineation
• Same framework / interface used in other SKUs • Built-in or Peripheral, Touch vs Swipe, Requirements • Improve resiliency to threats (recommendations) • Ensure the device / driver meets the security bar for publishing on Windows Update
• Driver built for x86 desktop can be recompiled to work on ARM • From IHV perspective, process of creating driver is same for all SKUs • Requires work with ODM to build BSPs • Some IHVs already have experience in this area (ex. FPC, Synaptics ) other’s don’t and so those IHVs will need to work closely with QC and the ODM to understand how to integrate their driver into a mobile BSP for Windows Phone
• Driver documentation and samples available on MSDN • Variety of options make this an easy path to enabling Windows Hello • Touch sensor recommended (better user experience) • No difference in implementation or requirements
• False Accept Rate (FAR) < 0.001% (large sensor) < 0.002% (swipe and small sensor) • False Reject Rate (FRR) < 5% without anti-spoofing • False Reject Rate (FRR) < 10% with anti-spoofing • Anti-spoofing solution is required • Full details available on MSDN and Connect • Size – 10mm x 10mm • Bus – SPI or USB (SPI preferred) • Power - <= 100mW during capture and <= 1mW otherwise • Better performance
• Spoofing and replay attacks • Injection of biometric samples • Template theft and injection • Attacks by privileged code on a compromised system • Protect biometric input, from raw data collection through template matching • When possible utilize “advanced sensors” capabilities to perform match on chip • Isolate biometric operations and template management with TrustZone • Implementation details available on Connect (V4 engine adapter interface)
• Partner to provide report on how FAR results were achieved • Partner to submit results via SysDev bug for security review by feature team • Provide sensor samples (~10) for self-host validation • Full details of security review can be found on Connect
Please provid vide e feedb dback ck on this s session ssion: aka.ms/ a.ms/wi winhecfeedback nhecfeedback
Recommend
More recommend