Facial Image Frequency Analysis for Accurate Liveness Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing liveness detection methods for biometric authentication are inconvenient and inaccurate, failing to reliably distinguish between live individuals and fraudulent biometric data, such as photographs, during remote transactions.
Innovation Solution
Employ multi-resolution convolution and octave convolution techniques to extract first and second frequency components from facial image data, comparing them against corresponding frequency maps generated from genuine data to determine matching criteria and similarity scores, using filters like high-pass Laplacian and low-pass Gaussian filters to identify fraudulent images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-resolution convolution and octave convolution techniques are used to extract frequency components, then liveness detection accuracy is improved, but device complexity increases
Solution Approach 1:
The patent segments the facial image into multiple frequency components (low-frequency and high-frequency) using multi-resolution convolution and octave convolution techniques. This segmentation allows the system to analyze different aspects of the image separately, improving liveness detection accuracy by examining texture, structure, and frequency characteristics that are difficult to replicate in fraudulent images.
Solution Approach 2:
The patent transforms the two-dimensional facial image into multiple frequency domains through convolution operations. By converting spatial frequency information into frequency space and analyzing images at different resolutions and scales, the system gains additional dimensional information that enhances detection capability while maintaining manageable processing complexity through structured analysis.
2Reliability
If frequency component extraction and comparison methods are implemented, then authentication reliability is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary frequency component extraction and creates frequency maps from genuine image data before actual authentication occurs. These pre-computed frequency maps serve as reference templates that can be quickly compared against captured images during authentication, reducing real-time processing requirements while maintaining high reliability through pre-established baseline comparisons.
Solution Approach 2:
The patent extracts and analyzes only the most discriminative frequency components (low-frequency and high-frequency) rather than processing every pixel and frequency detail. This selective extraction approach provides sufficient authentication reliability by focusing on the most telling frequency characteristics while significantly reducing processing time compared to exhaustive analysis.
Data Source
Figure 1~2
Figure 3~5
Figure 6
AI summary
A method for enhancing liveness detection of image data is provided that includes receiving image data of a person and using multi-resolution convolution and octave convolution techniques to extract first and second frequency components from the received image data. The method also includes comparing the first frequency component against data in a first frequency map and comparing the second frequency component against data in a second frequency map. Furthermore, the method includes determining whether the first frequency component satisfies matching criteria with data in the first frequency map and determining whether the second frequency component satisfies matching criteria with data in the second frequency map. In response to determining the first and second frequency components fail to satisfy the data in the respective frequency map, the method concludes that the received image data was not taken of a live person.