Facial Recognition Feature Fusion for Rapid Pose Changes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition technologies struggle with accuracy when faced with rapid and violent changes in facial pose, leading to a significant gap in performance in practical applications.
Innovation Solution
A facial recognition method that fuses facial pose features, including spatial information such as structural, edge, and angle features, with facial features like skin color and texture, to generate enhanced facial pose information and features, determining a user recognition result only when the pose meets preset conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing facial detection technologies are used, then good performance is achieved for faces in ideal environments, but poor environment generalization capability and performance effect occur in complex environments with rapid and violent facial pose changes
Solution Approach 1:
The patent segments the facial recognition process into two distinct feature extraction streams: facial pose feature extraction and facial feature extraction. This segmentation allows each stream to specialize in extracting specific types of information (pose-related spatial information vs. skin color and texture information), enabling the system to handle complex environments and rapid pose changes more effectively by processing different feature types independently before fusion.
Solution Approach 2:
The patent merges facial pose features and facial features through feature fusion to generate comprehensive recognition results. By combining pose features (structural, edge, and angle information) with facial features (skin color, texture, and other characteristics), the system achieves better environment generalization capability and maintains high recognition precision across varying conditions that existing single-feature methods cannot handle.
2Measurement precision
If facial recognition is performed without considering facial pose features, then simpler processing is achieved, but accuracy deteriorates when facial pose changes rapidly and violently
Solution Approach 1:
The patent segments feature extraction into specialized sub-tasks: one stream extracts facial pose features (structural, edge, angle information) while another extracts facial features (skin color, texture). This segmentation improves accuracy by ensuring pose-related information is explicitly captured, while the modular structure manages complexity through organized, independent processing streams.
Solution Approach 2:
The patent applies local quality by assigning different extraction focuses to different feature streams: pose feature extraction concentrates on spatial and geometric information, while facial feature extraction focuses on dermatological characteristics. This localized specialization ensures each stream optimizes for its specific purpose, improving overall recognition accuracy without requiring all components to handle all types of information.
Data Source
AI summary
A facial recognition method and apparatus are provided. In the method, facial pose features and facial features can be fused; that is, detailed information of the facial pose features and the facial features can be fused, and a target facial feature and facial pose information of a facial image to be subjected to recognition are generated according to the fused features, such that accuracy of the determined target facial feature and facial pose information is improved, thereby improving accuracy of a user recognition result determined according to the target facial feature and the facial pose information.