Multi-task Learning Network for Face Liveness Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image-based face anti-spoofing methods face challenges in detection accuracy and robustness against nuisance variables, particularly when only a single image is provided, and lack a segmented guided classification approach.
Innovation Solution
A novel multi-task learning network that combines segmentation and classification tasks, utilizing a face detection module, segmentation module, classification module, and fusion module to identify the liveliness of facial images by annotating pixel-level semantic information and determining probabilistic features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional single-task classification methods are used for face anti-spoofing, then the system complexity is low, but the detection accuracy and robustness are insufficient
Solution Approach 1:
The patent divides the face anti-spoofing task into two separate modules: a segmentation module that performs pixel-level classification to identify real vs. spoofed regions, and a classification module that provides global authentication decisions. This segmentation allows each module to specialize in its specific function, improving overall detection accuracy while maintaining manageable system complexity through modular design.
Solution Approach 2:
The segmentation module serves dual purposes: it provides detailed pixel-level authentication information and also generates features that feed into the classification module. This multi-functionality allows the system to leverage the same processed information for both local and global authentication decisions, improving robustness without proportionally increasing complexity.
2Reliability
If multiple spatial scales and additional information are leveraged, then detection robustness improves, but the processing complexity and computational load increase
Solution Approach 1:
The patent incorporates multi-scale processing by analyzing facial images at different spatial resolutions. The segmentation module processes features at multiple scales to capture both fine-grained pixel-level details and broader contextual patterns, enhancing robustness to variations in lighting, pose, and image quality without requiring separate processing pipelines for each scale.
3Measurement precision
If pixel-level semantic information is extracted through segmentation, then the detection precision is improved, but the computational time and resources increase
Solution Approach 1:
The segmentation module performs preliminary pixel-level analysis and generates segmentation maps that identify real and spoofed regions. These pre-processed segmentation results are then utilized by the classification module, avoiding redundant computations and enabling faster global authentication decisions based on already-analyzed pixel-level information.
Solution Approach 2:
The patent combines the segmentation and classification modules into an integrated multi-task learning framework where both modules share common feature extraction layers. This merging allows the system to simultaneously perform pixel-level segmentation and global classification using shared computational resources, reducing overall processing time compared to running separate independent models.
Data Source
AI summary
The present invention discloses a liveliness detection technique. The technique is described for identifying facial attributes. The technique identifies the presented face in the image as real or deceptive. The system and method includes identifying the facial attributes and utilizing a multi task learning network. The neural network includes segmentation and classification functionalities. The final output is used to get pixel level semantic information and high level semantic information.