Face Search Feature Mapping for Multi-Model Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Face search systems face challenges in selecting an appropriate feature space for comparing face features extracted from different models, leading to reduced accuracy and increased computational costs due to the need for extensive feature mapping and training of multiple models.
Innovation Solution
A method that involves concatenating features from multiple feature extraction models and using a feature mapping model to create a standard feature, allowing for comprehensive application of feature extraction models while reducing computational load by training fewer feature mapping models and selecting an appropriate feature space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple feature extraction models are comprehensively applied to improve face search accuracy, then the face search accuracy is improved, but the computational cost and device complexity increase exponentially
Solution Approach 1:
The patent segments the feature comparison task by introducing an intermediate feature mapping model. Instead of directly comparing features from multiple different extraction models (which would require exponential computational resources), the system divides the problem into: (1) extracting features using multiple models, (2) mapping each model's features to a unified feature space through dedicated mapping models, and (3) comparing features in the unified space. This segmentation reduces computational complexity from exponential to linear scale.
Solution Approach 2:
The patent introduces feature mapping models as intermediary components between different feature extraction models and the final comparison stage. These mapping models serve as mediators that transform features from various extraction models into a common feature space, enabling accurate comparisons without requiring direct pairwise comparisons between all model combinations. This intermediary approach maintains measurement precision while dramatically reducing device complexity.
2Adaptability or versatility
If features from different feature extraction models are mapped to the same feature space to enable direct comparison, then face features from different models can be compared, but the selection of appropriate feature space becomes a complex problem
Solution Approach 1:
The patent creates a universal feature space through the feature mapping model that can accommodate features from multiple different extraction models. Each mapping model is trained to transform its corresponding extraction model's features into this universal space, enabling any combination of feature extraction models to be compared without requiring separate comparison mechanisms for each pair. This universal approach enhances adaptability while managing complexity through standardized transformation processes.
3Adaptability or versatility
If all different face features are mapped to the same feature space to enable comprehensive model application, then multiple feature extraction models can be applied, but determining how to comprehensively apply multiple models becomes problematic
Solution Approach 1:
The patent segments the model application process by assigning each feature extraction model its dedicated feature mapping model. This segmentation allows each extraction model to be independently applied and mapped without requiring complex integration logic for combining multiple models. The systematic one-to-one correspondence between extraction models and mapping models simplifies the comprehensive application of multiple models while managing integration complexity.
Data Source
AI summary
A face search method and apparatus are provided. The method includes obtaining a to-be-searched face image, and inputting the face image into a first feature extraction model to obtain a first face feature. The method further includes inputting the face image and the first face feature into a first feature mapping model for feature mapping, to output a standard feature corresponding to the first face feature, and performing face search for the face image based on the standard feature. Features extracted by using a plurality of feature extraction models are concatenated, and a concatenated feature is used as a basis for constructing a standard feature.