Reinforced Palmprint Reconstruction Attacks in Biometric Systems.

Area of Science:

• Biometrics and Security Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Biometric systems are vulnerable to reconstruction attacks where leaked templates are used to create fake identities.
• Existing palmprint reconstruction attacks suffer from poor visual quality, unnaturalness, and incompleteness.

Purpose of the Study:

• To propose two novel, reinforced palmprint reconstruction attacks to overcome limitations of existing methods.
• To enhance the success rate and visual quality of biometric template reconstruction attacks.

Main Methods:

• Utilizing deep reinforcement strategies to iteratively modify palmprint regions of interest, reducing matching distance.
• Introducing Modification Constraint within Neighborhood (MCwN) to limit modification extent and prevent reckless changes.
• Implementing Batch Member Selection (BMS) to modify significant pixels (SPs) for reduced matching numbers and visual degradation.

Main Results:

• The proposed reinforced attacks achieve high success rates against state-of-the-art coding-based palmprint systems.
• Both MCwN and BMS attacks demonstrate improved naturalness and visual quality compared to prior methods.
• The attacks successfully satisfy requirements that existing methods cannot meet simultaneously.

Conclusions:

• The developed reinforced palmprint reconstruction attacks offer a significant advancement in biometric security threats.
• These methods provide a more effective and visually convincing means for attackers to compromise palmprint-based authentication systems.
• The findings highlight the need for enhanced defenses against sophisticated reconstruction attacks in biometric identity management.