Joyce Yang（NIST）：FATE Quality SIDD.pdf

1、NIST FATE Quality Specific Image Defect Detection(SIDD)Joyce Yang-NIST Image GroupIFPC 2025:Virtual Presentation,April 1 2025 FRTE/FATE Team:Patrick Grother,Mei Ngan,Kayee Hanaoka,Austin Hom,Joyce Yang,Jim MateyNIST Face BenchmarksFRTEFACE RECOGNITIONRECOGNITION TECHNOLOGY EVALUATIONFATEFACE ANALYSI

2、S ANALYSIS TECHNOLOGY EVALUATION1:1 VERIFICATION1:N SEARCHFACE IN VIDEO 2024 TWINS DISAMBIGUATIONMORPH DETECTIONQUALITY+DIAGNOSTICSPADAGE ESTIMATIONRECOGNITION:WHO IS IN AN IMAGEANALYSIS:WHAT ABOUT AN IMAGESAME PERSON OR NOT?WHO?WHERE?WHEN?1:N ON NON-COOP PEOPLESAME PERSON,OR TWIN?TWO PEOPLE IN ONE

3、FACE?HOW IS THIS PHOTO BAD?SUBVERSIVE PHOTO?HOW OLD?OLD ENOUGH?FRTE/FATE evaluations are:Why Specific Image Defect Detection?Automated face recognition is sensitive to quality problems in images Standards set requirements on images For MRTD,ISO/IEC:39794-5 For law enforcement,ANSI/NIST standard SIDD

4、 gives users a set of tools to check those requirementsUS Passport Photo Guidelines from https:/travel.state.gov/content/travel/en/passports/how-apply/photos.htmlWhy SIDD?Example:Passport PhotoFace capture software should be able to point out an issue such as the mouth being open and give feedback i

5、n real time.A SIDD algorithm might provide the output below.SubjectPoseYaw=20Face forwardMouthOpenness=0.08Close mouthBackgroundUniformity=0.8Change backgroundEyeGlassesPresent=1Remove eyeglassesFATE SIDD At A GlanceParticipation By YearFace CountMouth OpenFace OcclusionDistance to EdgesUnder-exposu

6、reOver-exposureResolutionUnified Quality ScoreMotion BlurCompression ArtifactsInter-Eye Distance(IED)Background UniformitySunglasses and EyeglassesEyes OpenPoseSubject-relatedCapture-relatedOtherThe two main goals of SIDD are to1)Promote quality defect detection algorithm development and 2)Support I