Biometric Verification Against Adversarial Facial Samples
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Biometric recognition systems are vulnerable to adversarial sample attacks, particularly in financial applications, which can compromise security and integrity by misguiding facial recognition models.
Innovation Solution
A biometric recognition method involving random transformation of low-bit data, pre-processing with filtering algorithms, and identity verification based on pre-processed data, along with the use of adversarial samples to counteract negative adversarial samples, and a blacklist rejection strategy to handle inconsistent results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If biometric recognition systems use facial image data for identity verification, then identity verification capability is improved, but vulnerability to adversarial sample attacks increases
Solution Approach 1:
The patent applies preliminary action by performing random transformation on low-bit data before the main identity verification process. This pre-processing step modifies the least significant bits of the biometric data in advance, creating a transformed version that will be used for verification. By preparing the data with random transformations beforehand, the system establishes multiple verification pathways that can withstand adversarial attacks during the actual verification process.
Solution Approach 2:
The patent implements parameter changes by randomly transforming specific bits (low-bit data) of the biometric information. This involves changing the state of certain data parameters (the least significant bits) from their original values to randomized values. This parameter transformation creates variability in the verification process, making it difficult for adversarial samples to consistently manipulate the recognition outcome.
2Reliability
If random transformation of low-bit data is performed, then resistance to adversarial attacks is improved, but data processing complexity increases
Solution Approach 1:
The patent applies local quality by focusing the random transformation operation specifically on the low-bit data (least significant bits) rather than the entire biometric dataset. This localized approach means that only a small portion of the data undergoes complex transformation, while the majority of the data remains unchanged. This reduces the overall computational burden and processing complexity while still providing the security benefits of randomization against adversarial attacks.
Solution Approach 2:
The patent implements partial action by applying random transformation to only a subset of the data bits (specifically the least significant bits) rather than transforming the entire dataset. This partial transformation is sufficient to provide security against adversarial attacks while avoiding the excessive computational complexity that would result from transforming all data. The selective application of transformation maintains efficiency while achieving the security objective.
3Measurement precision
If multiple filtering algorithms are used for pre-processing, then accuracy of identity verification is improved, but processing time increases
Solution Approach 1:
The patent applies dynamics by randomly selecting filtering algorithms for each verification process rather than using a fixed sequence of filters. This dynamic selection means that the processing path varies from one verification attempt to another, preventing adversarial samples from being optimized for a specific filter sequence. The randomness in algorithm selection adds adaptability to the system, improving accuracy through multiple filtering options while managing processing time through efficient random selection rather than exhaustive application of all possible filters.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
The disclosure pertains to a biometric recognition method, the method comprising: receiving biometric data; performing a random transformation of low-bit data in received biometric data; randomly selecting one or more filtering algorithms to pre-process randomly transformed biometric data; and performing identity verification based on pre-processed biometric data. Furthermore, the disclosure involves a method for generating an adversarial sample, a biometric recognition server, a client, a computer storage medium, and a computer program product.