Face Anti-Spoofing via Multi-View CNN Image Concatenation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face anti-spoofing methods either burden users with numerous interactions or fail to achieve accurate liveness detection due to reliance on single images and discarded views of the face.
Innovation Solution
A method involving a convolutional neural network (CNN) classifier that receives and concatenates multiple images of a user's face, each corresponding to different interaction challenges, to determine liveness by comparing a confidence score against a threshold value, improving user experience and detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple face images are captured with numerous user interactions (shaking head, opening mouth, blinking eye), then the liveness detection accuracy is improved, but the user experience deteriorates due to excessive interaction requirements
Solution Approach 1:
The patent applies partial action by capturing multiple face images from different views (front, left, right, top, bottom) but using a streamlined set of interactions compared to traditional methods. Instead of requiring numerous repeated interactions for each view, the system captures images more efficiently while still obtaining sufficient data for accurate liveness detection through the CNN model that processes multi-view image sequences.
2Ease of operation
If only one still face image is used for liveness detection, then the user interaction is minimized, but the detection accuracy deteriorates due to discarded images of other views
Solution Approach 1:
The patent transitions from analyzing a single 2D face image to processing multi-view 3D face images captured from different spatial angles (front, left, right, top, bottom views). This dimensional expansion allows the CNN model to analyze facial features from multiple perspectives simultaneously, significantly improving liveness detection accuracy by detecting spoofing attempts that might be invisible in a single view while maintaining minimal user interaction.
3Measurement precision
If multiple facial views are analyzed using CNN classifier, then the liveness detection accuracy is enhanced, but the computational complexity increases
Solution Approach 1:
The patent segments the face detection task into multiple independent view components (front view, left view, right view, top view, bottom view), each processed separately by the CNN classifier. This segmentation allows the system to handle complex multi-view analysis by breaking it down into manageable segments that can be processed in parallel or sequentially, reducing the overall computational burden while maintaining high detection accuracy through comprehensive multi-perspective analysis.
Data Source
AI summary
Disclosed are computer-implemented methods, non-transitory computer-readable media, and systems for face anti-spoofing. One computer-implemented method includes receiving, by a server and from a user device, a plurality of images of a user face, each of the plurality of images corresponding to a facial expression performed by a user in response to each interaction challenge of a plurality of interaction challenges. The plurality of images are concatenated into a concatenated image. A convolutional neural network (CNN) classifier is used to obtain a confidence score for the concatenated image. A liveness of the user face is determined based on a comparison of the confidence score and a threshold value.