3D-Aligned Face Recognition via Deep Neural Network Warping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current facial recognition systems face significant challenges in unconstrained environments due to factors like lighting, expression, occlusion, and aging, leading to decreased performance compared to human visual systems.
Innovation Solution
The use of a deep neural network (DNN) that aligns 2D face images to 3D shapes, generating 3D-aligned face images, which are then classified using convolutional-pooling, locally-connected, and fully-connected layers to extract and correlate features, improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional 2D face recognition systems are used in constrained environments, then recognition accuracy is high, but performance deteriorates substantially in unconstrained settings due to lighting, expression, occlusion, and aging
Solution Approach 1:
The patent transforms 2D face images into 3D-aligned representations by warping the 2D image to match a 3D face model. This dimensional transformation allows the system to account for non-planarity and non-rigid expressions, enabling reliable recognition in unconstrained environments where traditional 2D systems fail.
Solution Approach 2:
The system changes the parameter space from 2D pixel coordinates to 3D facial geometry parameters. By aligning 2D images to 3D face models and extracting features in this transformed parameter space, the system becomes invariant to variations in lighting, expression, and pose, thereby maintaining reliability across different conditions.
2Reliability
If 3D-aligned face images are generated from 2D face images using deep neural networks, then recognition accuracy in unconstrained settings improves, but system complexity increases
Solution Approach 1:
The patent replaces traditional mechanical/image processing alignment methods with a deep neural network that automatically learns and applies 3D warping transformations. The DNN takes 2D face images and directly outputs 3D-aligned face images, substituting complex manual alignment procedures with an automated learned transformation.
Solution Approach 2:
The deep neural network serves as an intermediary between 2D face images and the recognition system. It performs the complex task of 3D alignment and feature extraction, acting as a mediator that transforms raw 2D inputs into standardized 3D-aligned representations that can be reliably recognized regardless of input variations.
Data Source
AI summary
Systems, methods, and non-transitory computer readable media can align face images, classify face images, and verify face images by employing a deep neural network (DNN). A 3D-aligned face image can be generated from a 2D face image. An identity of the 2D face image can be classified based on provision of the 3D-aligned face image to the DNN. The identity of the 2D face image can comprise a feature vector.