Dual-Sensor Face Liveness Detection for Spoofing Under Variable Light
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face liveness detection methods struggle with low accuracy, particularly in recognizing spoofing faces, leading to security risks and personal information vulnerabilities.
Innovation Solution
A method involving two sensors to capture target images, extracting depth information using a pre-trained deep generative network, and employing a pre-trained liveness detection model to determine liveness based on this depth information, trained with sample data under varying lighting conditions to enhance detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional face liveness detection methods are used, then the detection process is simple, but the detection accuracy is low and spoofing faces cannot be reliably distinguished
Solution Approach 1:
The patent introduces depth information as a new dimension beyond traditional 2D image analysis. By capturing images from multiple sensors at different positions and extracting depth maps, the system transforms planar face images into three-dimensional representations, enabling more accurate distinction between live and spoofing faces through depth feature analysis
Solution Approach 2:
The patent divides the detection system into multiple independent components: multiple image sensors for capturing images from different positions, a deep generative network for depth information extraction, and a liveness detection model for final classification. This segmentation allows each component to specialize in specific tasks, improving overall detection accuracy while maintaining modular system architecture
2Measurement precision
If multiple sensors and deep learning models are used to improve detection accuracy, then liveness detection precision increases, but computational resource consumption increases
Solution Approach 1:
The patent pre-trains the deep generative network and liveness detection model using extensive sample data including various spoofing face materials and lighting conditions. This preliminary training enables the models to extract depth information and perform detection more efficiently during actual use, reducing real-time computational burden while maintaining high accuracy
Solution Approach 2:
The patent extracts only the essential depth information from the captured images using the pre-trained deep generative network, rather than processing entire high-resolution images. This extraction of key features reduces the data volume requiring further processing, thereby lowering computational resource consumption in the detection phase
3Adaptability or versatility
If training data from multiple lighting environments is used, then the liveness detection model becomes more robust to lighting variations, but the training process becomes more complex and time-consuming
Solution Approach 1:
The patent collects and prepares training data from multiple lighting environments in advance, including various spoofing face materials and lighting conditions. The deep generative network and liveness detection model are pre-trained on this diverse dataset before deployment. This preliminary preparation ensures the models learn to handle lighting variations efficiently, reducing the need for extensive real-time adjustments
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Disclosed in the present application are a liveness detection method and apparatus, a training method and apparatus for a liveness detection system, and an electronic device and a storage medium. The liveness detection method comprises: acquiring two target images, which are obtained by means of a first sensor and a second sensor respectively performing collection on the same face to be recognized; extracting target depth information from the two target images by using a deep generative network; and performing detection on the target depth information by using a liveness detection model, so as to obtain a liveness detection result of said face. The liveness detection model is obtained by means of training using depth information of sample data, the sample data comprises first sample images, which are collected by the first sensor under at least two types of illumination, and second sample images, which are collected by the second sensor under the at least two types of illumination, and each of the first sample images and each of the second sample images comprise spoofing faces of different materials. Target depth information is extracted from two images by means of a deep generative network, and liveness detection is performed on the target depth information by using a liveness detection model, so as to recognize spoofing faces of different materials under different types of illumination, such that the accuracy of recognition is higher.