Monocular Depth Estimation for Liveness Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing liveness detection systems are vulnerable to spoofing attacks and require complex hardware setups, limiting their usability and robustness, especially in mobile devices, and often fail to detect non-live attacks like presentation and mask-based attacks due to environmental factors and reliance on specific image information.
Innovation Solution
A method and system that captures images, processes them using a face detector to create target images, generates depth maps, and uses multi-branch image liveness models to detect non-live attacks by integrating depth and color models, enabling robust and accurate liveness detection on any device with a camera, including mobile devices, without requiring specific hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If passive liveness solutions use depth information from stereovision cameras or IR sensors, then liveness detection accuracy is improved, but hardware complexity and device requirements increase
Solution Approach 1:
The patent uses a monocular depth estimation model to generate depth maps from standard 2D images, copying the depth information that would normally require complex stereovision or IR sensors. This allows the system to obtain depth data using only a standard camera, eliminating the need for specialized hardware while maintaining detection accuracy
Solution Approach 2:
The patent replaces physical depth-sensing hardware (stereovision cameras, IR sensors) with a computational approach using a monocular depth estimation neural network. This substitutes mechanical/optical depth measurement systems with an algorithmic solution that processes standard 2D images to infer depth information
2Productivity
If face detection module crops only the face region for analysis, then processing speed is improved, but detection robustness against presentation and mask attacks deteriorates
Solution Approach 1:
The patent extends the analysis from the 2D face region to the full image by incorporating depth maps and analyzing the entire image area. This additional dimensional information (depth + full image context) allows detection of presentation attacks and masks that would be invisible if only the cropped face region were analyzed
Solution Approach 2:
The patent performs preliminary analysis on the full image and depth map before final face region analysis. By first examining the complete image context including background and depth information, the system can detect suspicious patterns (like presentation screens or mask edges) that would be missed if face cropping occurred first
3Reliability
If liveness detection systems use multiple sensors and complex hardware setups, then detection accuracy is improved, but ease of operation and accessibility deteriorates
Solution Approach 1:
The patent creates a universal liveness detection solution that works on any device with a standard camera. By using a monocular depth estimation model that processes regular 2D images, the system achieves sensor-independent operation, making it accessible on mobile phones, tablets, and computers without specialized hardware
Solution Approach 2:
The patent copies depth information from standard 2D images through computational methods, allowing the system to function on devices with only basic cameras. This eliminates the need for users to have access to expensive or specialized sensing equipment, greatly improving accessibility
Data Source
AI summary
The present invention relates to a method and system of image processing for determining liveness of a subject. The method comprises processing a captured image to create a plurality of target images. The method then encompasses generating a depth map corresponding to each target image from the plurality of target images. Further, the method comprises creating a plurality of modified images based on an addition of the depth map and a set of color models associated with the plurality of target images. Next, the method comprises detecting, by a plurality of multi-branch image liveness models, one of a presence of a set of non-live attacks and an absence of the set of non-live attacks in the plurality of modified images. Further the method leads to determining the liveness of the subject based on detection of the absence of the set of non-live attacks.