3D Facial Model Deformation for Pose-Invariant Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Facial recognition technology is sensitive to facial pose, expression, occlusion, and changes in illumination, which affects its accuracy and reliability in various applications.
Innovation Solution
A method and apparatus that extract facial landmarks from 2D images to adjust and match a deformable 3D facial model, projecting 3D feature points onto 2D images to extract image features, which are then compared to registered features for recognition, improving pose and expression adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If facial recognition uses traditional 2D image matching, then the system is simple to implement, but recognition accuracy deteriorates under variations in pose and expression
Solution Approach 1:
The patent transforms 2D facial recognition into 3D space by constructing a deformable 3D facial model. Facial landmarks extracted from 2D images are used to adjust and match the 3D model, allowing the system to capture depth information and spatial relationships. This dimensional transition enables accurate recognition under various poses and expressions while maintaining reasonable system complexity through efficient model deformation techniques.
Solution Approach 2:
The patent employs a deformable 3D facial model that can dynamically adapt to different facial poses and expressions. The model is adjusted by matching facial landmarks from input images to corresponding points on the 3D model, allowing real-time deformation to match the subject's current facial state. This dynamic adaptation maintains high recognition accuracy across varying conditions without requiring multiple static models.
2Reliability
If the system adapts to multiple poses and expressions using complex modeling, then recognition accuracy improves, but processing time increases
Solution Approach 1:
The patent pre-constructs a standardized 3D facial model with defined geometric features and landmarks before actual recognition occurs. This preliminary preparation allows the system to quickly deform and match the model to input images without performing complex computations during recognition. The pre-established model structure enables rapid adaptation to different poses and expressions, reducing processing time while maintaining accuracy.
Solution Approach 2:
The patent divides the facial recognition process into discrete landmark points and regions of interest on the 3D model. By focusing computation on matching specific facial landmarks (eyes, nose, mouth, chin) rather than processing the entire face uniformly, the system achieves accurate pose and expression adaptation with reduced computational overhead. This segmented approach allows parallel processing of multiple landmark matches, improving efficiency.
3Adaptability or versatility
If facial landmarks are extracted and used to adjust 3D model, then adaptability to pose and expression improves, but computational complexity increases
Solution Approach 1:
The patent introduces facial landmarks as intermediary points that bridge 2D image input and 3D model adjustment. Instead of directly comparing complex 2D images or manipulating the entire 3D model, the system uses a limited set of key landmark points (typically 68 or fewer) as mediators. These landmarks provide sufficient information to determine pose and expression while keeping computational requirements manageable. The landmarks serve as a simplified representation that captures essential facial geometry for model deformation.
Data Source
AI summary
A method and an apparatus for registering a face, and a method and an apparatus for recognizing a face are disclosed, in which a face registering apparatus may change a stored three-dimensional (3D) facial model to an individualized 3D facial model based on facial landmarks extracted from two-dimensional (2D) face images, match the individualized 3D facial model to a current 2D face image of the 2D face images, and extract an image feature of the current 2D face image from regions in the current 2D face image to which 3D feature points of the individualized 3D facial model are projected, and a face recognizing apparatus may perform facial recognition based on image features of the 2D face images extracted by the face registering apparatus.