DeepID2 Feature Extraction for Face Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition methods, particularly those with linear or shallow structures, are limited in addressing the complex, highly nonlinear intra- and inter-personal variations in high-dimensional image spaces, leading to suboptimal performance in face verification tasks.
Innovation Solution
The use of deep convolutional feature extraction systems that learn DeepID2 features through hierarchical nonlinear mappings by simultaneously incorporating identification and verification supervisory signals, which enhance inter-personal variations and reduce intra-personal variations, respectively, and extract complementary features from various face regions concatenated after PCA dimension reduction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If linear or shallow feature extraction models are used, then the system complexity is low, but the ability to capture complex nonlinear variations in high-dimensional image space is insufficient
Solution Approach 1:
The patent transitions from linear/shallow feature extraction to deep hierarchical feature extraction by adding multiple layers of convolutional and fully-connected layers. This dimensional transformation in the feature space enables the model to capture complex nonlinear variations in high-dimensional image space, directly resolving the contradiction between model complexity and verification accuracy.
Solution Approach 2:
The patent implements a nested hierarchical structure where multiple convolutional layers and fully-connected layers are stacked, with each layer extracting features at different levels of abstraction. The DeepID2 features are extracted through this nested structure, allowing the system to progressively capture both fine-grained and coarse-grained facial variations, thereby improving accuracy while managing complexity through structured organization.
2Measurement precision
If deep convolutional feature extraction systems with hierarchical nonlinear mappings are used, then face verification accuracy is improved, but the computational complexity and training difficulty increase
Solution Approach 1:
The patent merges the identification task (classification) and verification task (similarity comparison) into a single unified deep convolutional feature extraction system. By simultaneously incorporating identification and verification supervisory signals during training, the model learns features that are optimized for both tasks, reducing the need for separate systems and simplifying the overall training process despite the increased depth of the network.
Solution Approach 2:
The deep convolutional feature extraction system is designed to serve multiple functions: it extracts features for both identification (classification into specific identities) and verification (comparison of similarity). The same hierarchical nonlinear mappings are used to generate features that satisfy both identification accuracy and verification reliability, making the system multi-functional and reducing overall system complexity.
3Measurement precision
If identification and verification supervisory signals are incorporated simultaneously, then complementary features are learned that improve recognition, but the training process becomes more complex
Solution Approach 1:
The patent combines identification supervisory signals (class labels for classification) and verification supervisory signals (similarity targets for comparison) into a unified training objective. The loss function integrates both identification loss and verification loss, allowing the model to learn features that simultaneously satisfy both tasks. This merging approach, while increasing training complexity, produces complementary features that significantly improve recognition performance.
Data Source
AI summary
Disclosed is an apparatus for face verification. The apparatus may comprise a feature extraction unit and a verification unit. In one embodiment, the feature extraction unit comprises a plurality of convolutional feature extraction systems trained with different face training set, wherein each of systems comprises: a plurality of cascaded convolutional, pooling, locally-connected, and fully-connected feature extraction units configured to extract facial features for face verification from face regions of face images; wherein an output unit of the unit cascade, which could be a fully-connected unit in one embodiment of the present application, is connected to at least one of previous convolutional, pooling, locally-connected, or fully-connected units, and is configured to extract facial features (referred to as deep identification-verification features or DeepID2) for face verification from the facial features in the connected units. The verification unit may be configured to compare the obtained DeepID2 extracted from two face images to be compared to determine if the two face images are from the same identity or not.