Liveness Image Patch Analysis Without Resizing Distortion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques for liveness detection in biometric identification are prone to losing important features due to image resizing and distortion, leading to undetected spoofing attacks, and there is a need for systems and methods to efficiently detect liveness while preserving original imagery data characteristics.
Innovation Solution
The system splits imagery data into sets of patches, encodes and decodes them, and uses machine learning models to analyze each patch individually, aggregating results to determine liveness, with training pipelines aligned to maintain data integrity and handle unknown attacks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If image processing techniques are used to prepare imagery data for analysis models, then the imagery data can be processed by machine learning models, but important features characterizing spoofing attacks are lost due to resizing and distortion
Solution Approach 1:
The system segments the image processing task by analyzing imagery data at multiple different sizes simultaneously rather than resizing to a single fixed dimension. Each analysis model processes the same imagery data at its native resolution or multiple resolutions, preserving all important features while enabling model-specific optimizations.
Solution Approach 2:
The system changes the parameter of image resolution from a fixed single value to multiple variable values. Instead of converting all imagery data to one standard size, the system maintains and processes images at their original dimensions and additional sizes, allowing each analysis model to operate on optimally-sized data without feature loss.
2Productivity
If image resizing is performed to meet model input requirements, then machine learning models can effectively analyze the imagery data, but the integrity of the image is compromised
Solution Approach 1:
The system divides the analysis workflow into separate processing streams, where each analysis model receives imagery data in its own optimal format. Rather than forcing a single resizing operation, the system segments the data distribution so that each model gets appropriately-sized images without compromising the original image integrity for detection purposes.
Solution Approach 2:
The system performs preliminary preparation by creating and storing multiple versions of imagery data at different sizes before analysis begins. This preliminary action ensures that when analysis models need specific dimensions, the appropriately-sized versions are already available, eliminating the need for destructive resizing during the actual analysis phase.
3Productivity
If fixed matrix size input is required for image analysis models, then models can process images efficiently, but original image characteristics are distorted
Solution Approach 1:
The system creates a universal image processing framework that serves multiple analysis models with different requirements simultaneously. By generating and distributing imagery data at multiple sizes, the system makes the same input data universally usable across different models with different matrix size requirements, without forcing any single model to accept distorted or improperly-sized input.
Data Source
AI summary
Systems and methods for detecting liveness of imagery data. The imagery data, i.e., an image, is captured by an imagery application installed on a smartphone. The image is encoded and transferred to the liveness detector. At the liveness detector, the image is decoded and split into a set of patches by an image processor. Each of the sets of patches is analyzed. A liveness detector analyzes each patch of the set individually to determine if the patch is live or fake. By aggregating analysis results of all patches, liveness of the image is detected.