Frequency Filtering for Identity Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Facial recognition systems using CNNs often fail to authenticate users due to mismatches caused by cicatrix, marks, or changes in facial images, leading to premature authentication failures and the need for additional credentials or reauthentication.
Innovation Solution
A system that transforms facial images into a frequency domain, uses CNNs to determine image embeddings with high and low frequency components, and implements frequency filtering to mute mismatches, allowing for accurate authentication by classifying images and triggering access protocols.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If facial recognition systems use CNNs to compare input images with user-defined templates, then authentication capability is provided, but mismatches caused by cicatrix, marks, or changes lead to false negatives and authentication failures
Solution Approach 1:
The patent segments the facial image into multiple frequency components using Fourier Transform, separating the image into high-frequency components (edges, details, marks) and low-frequency components (overall structure, shape). This segmentation allows the system to process and compare different frequency components separately, reducing the impact of localized mismatches caused by cicatrix or marks on the overall authentication accuracy.
Solution Approach 2:
The patent changes the parameter domain from spatial domain to frequency domain by applying Fourier Transform. This parameter transformation allows the system to analyze and compare facial features in the frequency domain, where certain artifacts and marks appear as distinct high-frequency components that can be filtered or weighted differently, thereby improving measurement precision while maintaining reliability.
2Adaptability or versatility
If tolerance levels are built into recognition models to accommodate facial variations, then more facial changes are accepted, but regions with cicatrix or marks still cause mismatches and authentication failures
Solution Approach 1:
The patent applies local quality by treating different frequency components with different importance weights. Low-frequency components (representing overall facial structure) are given higher weight for matching, while high-frequency components (representing local variations like marks and cicatrix) are given lower weight or subjected to different processing. This allows the system to be more adaptable to facial variations while maintaining reliable authentication by focusing on stable, distinctive features.
Solution Approach 2:
The patent introduces another dimension (frequency dimension) to the traditional spatial domain image comparison. By transforming images into the frequency domain and comparing them there, the system gains an additional perspective for analyzing facial similarities and differences, enabling better handling of tolerances for facial variations while maintaining authentication reliability.
3Reliability
If strict matching criteria are used to ensure security, then false positives are reduced, but users with facial changes experience authentication failures and require additional credentials
Solution Approach 1:
The patent implements dynamic processing by applying different processing strategies to different frequency components. The system dynamically adjusts the matching process by filtering, weighting, or separately processing high-frequency and low-frequency components based on their characteristics. This dynamic approach allows the system to maintain strict security criteria for essential features while being more flexible with transient variations, thereby improving ease of operation without compromising security.
Data Source
AI summary
Systems, computer program products, and methods are described herein for frequency filtering in image analysis for identity verification. The present invention is configured to receive, from a user input device, a request from a user to initiate identity verification for access privileges; receive a first image of a face of the user from the user input device in response to the request, wherein the first image is in a geometric domain; transform, using an image transformation algorithm, the first image from the geometric domain into a frequency domain; process, using a machine learning (ML) subsystem, the first image of the user; authenticate the user based on at least processing the first image; and automatically trigger one or more access privilege protocols in response to authenticating the user.