Evasion of Facial Recognition Systems via Adversarial Makeup

Key Highlights & Summary 

• Anti-Surveillance Tactics: Recent public discussions highlight using cosmetics, painted facial features, and high-contrast patterns to scramble computer vision algorithms during public gatherings.  

• Working Mechanism of FRS: Facial Recognition Systems (FRS) utilize artificial intelligence models to convert facial features into numerical face embeddings to match against database records.  

• Adversarial Makeup Research: Historically inspired by WWI military camouflage (Dazzle), projects like Adam Harvey 2010 CV Dazzle and 2021 Ben-Gurion University experiments proved that machine-learning-designed makeup can bypass state-of-the-art FRS.  

• Practical Limitations: Evasion techniques designed for machine algorithms often fail under varying light conditions or high-resolution cameras and leave individuals highly conspicuous to human observers. 

• Regulatory Exemption Gaps: In India, public surveillance deployment operates without a dedicated statutory regulatory framework governing facial recognition use by law enforcement agencies.  

• Privacy and Data Safeguards: Facial images and numerical digital templates qualify as personal biometric data under domestic data protection laws, raising significant civil liberty concerns.  

Essential Definitions

• Facial Recognition System (FRS): A biometric technology that maps visual facial characteristics using AI models to create numerical face embeddings for identity verification or surveillance.  

• Face Embedding: A mathematical or numerical vector representation of an individual\'s unique facial geography generated by deep learning models.  

• Adversarial Attack: Technique in artificial intelligence where subtle modifications are made to input data (such as facial makeup) to cause machine learning models to make false predictions. 

Legal and Constitutional Framework 

• Digital Personal Data Protection (DPDP) Act, 2023: Classifies facial images and digital templates as personal data; however, Section 17 grants state instrumentalities broad exemptions for law enforcement, crime prevention, and public order maintenance.  

• Puttaswamy Judgment (2017): The Supreme Court declared privacy a Fundamental Right under Article 21, establishing the three-fold test of legality, legitimate state aim, and proportionality for state surveillance. 

• Information Technology Act, 2000: Governs digital data storage and processing, including rules for reasonable security practices for sensitive personal data.

• Article 21 of the Constitution: Protects individual autonomy, bodily privacy, and right against arbitrary state surveillance without proportional legislative safeguards. 

Conclusion 

While software-generated makeup patterns can occasionally disrupt FRS identification, tech variability limits its real-world reliability. Addressing AI surveillance risks requires robust statutory frameworks that balance public order imperatives with constitutional privacy rights.  

UPSC Relevance 

• Syllabus Connection: GS Paper-II (Governance, Fundamental Rights, Statutory Frameworks) & GS Paper-III (Internal Security, AI & Cyber Security).

 • Prelims Takeaway: DPDP Act 2023 provisions (Section 17 exemptions), Puttaswamy doctrine, and basic mechanics of FRS/face embeddings.