Face Image Extraction Using Depth-Checked Multispectral Liveness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition systems struggle to accurately distinguish between living and non-living bodies, particularly for individuals with deep skin tones, and infrared-based systems suffer from reduced accuracy compared to visible light images.
Innovation Solution
An image processing device and method that combines visible light and near-infrared images with depth information to detect and extract face images from living bodies, using a camera and projector to capture and analyze images, and a liveness check to determine if the face is from a living person.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If infrared light is used for face recognition, then authentication security is improved, but recognition accuracy deteriorates for individuals with deep skin tones
Solution Approach 1:
The patent combines infrared images and visible light images into a multi-spectral image data structure. The infrared image provides security against impersonation while the visible light image compensates for poor infrared reflection from deep skin tones. By merging both image types and using depth information to verify liveness, the system achieves both high security and high accuracy for diverse skin tones.
2Reliability
If infrared imaging is used to detect living bodies, then impersonation prevention is improved, but detection accuracy for deep colored skin deteriorates
Solution Approach 1:
The patent introduces depth information as an intermediary verification mechanism. Instead of relying solely on infrared reflection (which fails for deep skin tones), the system uses depth data to confirm that the detected face corresponds to a three-dimensional living object. This intermediary verification step ensures accurate detection while maintaining impersonation prevention through infrared imaging.
3Device complexity
If only infrared images are processed, then system complexity is reduced, but face recognition accuracy deteriorates
Solution Approach 1:
The patent segments the image processing into distinct functional components: acquiring both infrared and visible light images, extracting depth information, detecting faces in the multi-spectral data, and verifying liveness using depth cues. This segmentation allows each component to handle specific tasks efficiently, improving overall accuracy while keeping the system architecture manageable through modular processing stages.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Accurately determines living bodies and extracts face images suitable for face recognition, improving system accuracy and detecting potential dishonest acts.
Implementation Method 1
When a person having a deep colored skin such as a black person is captured by an infrared camera, however, a reflected infrared light cannot be sufficiently detected due to a large absorption rate of the infrared light
Implementation Method 2
using a camera and projector to capture and analyze images
Data Source
AI summary
An example embodiment includes: an acquisition unit that acquires a first image generated by capturing an object by using a light at a first wavelength, a second image generated by capturing the object by using a light at a second wavelength, and depth information on the object; a detection unit that detects a face included in the second image; a check unit that, based on the depth information, checks whether or not a face detected by the detection unit is one obtained by capturing a living body; and an extraction unit that, based on information on a face checked by the check unit as obtained by capturing a living body, extracts a face image from the first image.