3D Face Model Generation for Pose-Invariant Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face recognition technology is sensitive to face pose, facial expressions, occlusion, and changes in illumination, which affects its accuracy in recognizing users from images.
Innovation Solution
A method that generates a personalized 3D face model from a 2D input image, using deep neural networks to extract feature information from both pixel color values and 3D shape data, and normalizes the image to improve matching with enrolled user faces, enabling robust face recognition across varying conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional 2D face recognition is used, then the system is simple to implement, but accuracy deteriorates under varying lighting, pose, and occlusion conditions
Solution Approach 1:
The patent transforms 2D face images into 3D face models by inferring depth information and generating three-dimensional geometric representations. This dimensional transition enables the system to capture facial structure and topology that are invariant to lighting and pose variations, thereby improving recognition reliability while managing complexity through automated 3D reconstruction algorithms
Solution Approach 2:
The patent extracts and utilizes multiple parameters from 3D face models including surface normals, curvature information, and geometric depth maps. By transforming the face representation from 2D pixel values to 3D geometric parameters, the system achieves robustness against illumination changes and pose variations while maintaining a manageable complexity through parameterized 3D models
2Adaptability or versatility
If 3D shape information is incorporated, then robustness to lighting and pose changes improves, but processing complexity increases
Solution Approach 1:
The patent segments the face recognition process into distinct stages: 2D image acquisition, 3D model generation, normalization to canonical pose, feature extraction from 3D geometry, and matching. This segmentation allows each module to be optimized independently, managing overall processing complexity while achieving robustness through 3D shape analysis
Solution Approach 2:
The patent performs preliminary normalization of 3D face models to a canonical pose and lighting condition before feature extraction and matching. By pre-aligning and pre-processing the 3D representations, the system reduces the complexity of handling pose and illumination variations during the actual recognition process, thereby improving adaptability while controlling processing complexity
Data Source
AI summary
Face recognition of a face, to determine whether the face correlates with an enrolled face, may include generating a personalized three-dimensional (3D) face model based on a two-dimensional (2D) input image of the face, acquiring 3D shape information and a normalized 2D input image of the face based on the personalized 3D face model, generating feature information based on the 3D shape information and pixel color values of the normalized 2D input image, and comparing the feature information with feature information associated with the enrolled face. The feature information may include first and second feature information generated based on applying first and second deep neural network models to the pixel color values of the normalized 2D input image and the 3D shape information, respectively. The personalized 3D face model may be generated based on transforming a generic 3D face model based on landmarks detected in the 2D input image.