3D Facial Pose Correction for Large-Angle Face Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computerized facial recognition systems face difficulties in accurately comparing facial images with large pose angles to those with small pose angles due to computational challenges and potential image degradation during rotation.
Innovation Solution
A method and system utilizing machine learning models to generate multiple variations of a captured image at large poses by constructing base and captured facial objects, applying selected angles to rotate these objects, and comparing them with base images using an image matching model to determine a match within a threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If facial images with large pose angles are directly compared to small pose images, then identification speed is maintained, but recognition accuracy deteriorates due to computational difficulties and image degradation
Solution Approach 1:
The system performs preliminary pose correction on captured facial images before comparison. A pose correction module adjusts large pose images to align with the standard small pose orientation, ensuring that facial features are properly oriented before matching. This preliminary alignment improves recognition accuracy by eliminating pose-related mismatches while maintaining efficient comparison operations.
Solution Approach 2:
The patent introduces an intermediate pose correction step as a mediator between image capture and comparison. The pose correction module acts as an intermediary that transforms captured images into a standardized pose format, enabling accurate comparison without directly comparing disparate pose angles. This intermediary process resolves the contradiction by preparing images in advance for optimal matching.
2Measurement precision
If multiple pose angle variations are generated and compared, then recognition accuracy for large pose images improves, but processing time increases
Solution Approach 1:
The system pre-generates multiple pose angle variations of reference facial images during an offline preparation phase. These pre-computed pose variations are stored for rapid retrieval and comparison during actual recognition operations. By performing the computationally intensive pose variation generation in advance, the system achieves high recognition accuracy for large pose images without adding significant processing time during real-time operations.
Solution Approach 2:
The patent implements a dynamic comparison strategy where the system adaptively selects which pre-generated pose variations to use based on the captured image's pose characteristics. Rather than comparing against all possible pose variations, the system dynamically identifies and compares only the most relevant pre-generated poses, reducing processing time while maintaining high accuracy for large pose recognition.
Data Source
Figure 1A
Figure 1B
Figure 2A
AI summary
A method including generating a captured facial object and a captured pose from a captured image. The method also includes obtaining a base facial object and a base pose from a base image. The method also includes generating base pose angles using the captured pose, and captured pose angles using the captured pose. The method also includes obtaining selected base images using the base pose angles and the base facial object. The method also includes generating selected captured images using the captured pose angles and the captured facial object. The method also includes comparing the selected base images to the selected captured images to establish a comparison. The method also includes outputting a match output using the comparison.