Biometric Face Pose Verification via Landmark Density Ratios
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining biometric passport photos are inaccurate, especially for individuals with highly asymmetrical faces, as they rely on symmetry calculations that fail to precisely control the pose within the required tolerances for ICAO standards.
Innovation Solution
A method that determines a sequence of facial images from a video recording, calculates landmark densities across sectors of the face, and checks if the coordinate point representing the face orientation lies within a tolerance range of a reference point unique to the individual, iteratively adjusting the reference point to ensure accurate head orientation alignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If symmetry calculation is used for pose verification, then the process is simple, but the accuracy is insufficient especially for highly asymmetrical faces
Solution Approach 1:
The face is divided into multiple sectors using boundary lines that pass through predefined biometric landmarks. This segmentation allows for localized density analysis in different regions of the face, providing more granular pose verification data than global symmetry calculation.
Solution Approach 2:
Instead of treating the entire face uniformly, the method calculates landmark density for each sector individually. This local quality approach enables the system to detect pose deviations in specific facial regions, improving accuracy for asymmetrical faces where global symmetry metrics fail.
2Use of energy by moving object
If traditional symmetry check is used, then computational resources are saved, but pose verification accuracy deteriorates for asymmetrical faces
Solution Approach 1:
The face is divided into multiple sectors using boundary lines that pass through predefined biometric landmarks. This segmentation allows for localized density analysis in different regions of the face, providing more granular pose verification data than global symmetry calculation.
Solution Approach 2:
Instead of treating the entire face uniformly, the method calculates landmark density for each sector individually. This local quality approach enables the system to detect pose deviations in specific facial regions, improving accuracy for asymmetrical faces where global symmetry metrics fail.
3Reliability
If individual images are captured and re-captured until test criteria are met, then image quality can be ensured, but time consumption increases
Solution Approach 1:
The system performs preliminary pose verification using landmark density ratios on individual images before committing to capture sequences. This preliminary check allows the system to quickly identify suitable images and avoid unnecessary re-captures, reducing time loss while maintaining quality standards.
4Productivity
If video image recording is used to capture sequence of facial images, then productivity improves, but complexity of determining ideal pose increases
Solution Approach 1:
The face is divided into multiple sectors using boundary lines that pass through predefined biometric landmarks. This segmentation allows for localized density analysis in different regions of the face, providing more granular pose verification data than global symmetry calculation.
Solution Approach 2:
The system calculates landmark density ratios for opposite sectors and uses these ratios as feedback to evaluate pose quality. This quantitative feedback mechanism automatically identifies the ideal pose from the video sequence without requiring complex manual analysis, balancing productivity gains with manageable process complexity.
Data Source
Figure 1
Figure 2
Figure 3a~3c
AI summary
A method for determining an image data set for a biometric passport photo comprises determining a sequence of facial images for a person from a video image recording and, for each of the facial images, determining a plurality of predefined biometric landmarks in the facial image; dividing the facial image into several sectors, wherein the division is carried out by means of at least one boundary line dividing the facial image, which passes through a predefined selection of the biometric landmarks; determining a landmark density for each of the sectors, wherein the landmark density corresponds to the ratio of the number of landmarks located in the respective sector to the area of the sector; determining a density ratio between landmark densities that have been determined for sectors opposite each other within the facial image;and determining a coordinate point (3, 4, 5, 6, 7) for the facial image, where one coordinate value of the coordinate point (3, 4, 5, 6, 7) corresponds to the density ratio. A current facial image is selected from the sequence of facial images, and a pose check is performed for the current facial image. The pose check includes determining whether the coordinate point (3, 4, 5, 6, 7) lies within a tolerance range around a reference point, where the reference point represents a point unique to the individual, corresponding to a desired head orientation for a biometric passport photo; if the coordinate point is not within the tolerance range, selecting the next current facial image for the individual from the video image recording and performing the pose check for the next current facial image;and if the coordinate point lies within the tolerance range, the current facial image is determined as the selected facial image. The image data set is generated using the selected facial image. Furthermore, a method for personalizing a security document, a device for determining an image data set for a biometric passport photo, and a computer program product have been created.