Facial Recognition with Region Segmentation and Weighted Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition technologies face significant accuracy issues when the face is partially or fully covered, such as by sunglasses or masks, leading to decreased recognition performance.
Innovation Solution
A method that involves image segmentation to distinguish between covered and non-covered regions of a facial image, calculating weights for each region, and using pre-trained face feature extraction networks to generate feature vectors for comparison with template images, thereby improving recognition accuracy by reducing the impact of obstructions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional facial recognition methods are used on covered faces, then the recognition process can be completed, but the recognition accuracy decreases sharply
Solution Approach 1:
The patent divides the facial image into multiple regions based on coverage status. The image processing unit segments the face image into a first region (covered area) and a second region (uncovered area), allowing differential processing of different facial regions to improve recognition accuracy despite partial occlusion.
Solution Approach 2:
The patent applies different processing strategies to different regions of the face image. The feature extraction unit extracts features from both covered and uncovered regions using appropriate methods for each region type, and the determination unit combines these features with region-specific weights to achieve accurate recognition while accounting for local variations in quality and reliability.
2Reliability
If the face is covered by a large area (sunglasses, masks), then the global feature information is interfered, but the facial recognition process must still be completed
Solution Approach 1:
The patent segments the facial image into covered and uncovered regions, extracting features from each region separately. This allows the system to preserve useful local feature information from uncovered areas while appropriately handling the interfered global features from covered areas, thereby maintaining recognition reliability despite information loss.
Solution Approach 2:
The patent introduces region-based weighting parameters to adjust the contribution of different facial regions to the overall recognition decision. By dynamically adjusting these parameters based on coverage status, the system compensates for information loss in covered regions and emphasizes reliable features from uncovered regions, maintaining recognition performance.
Data Source
AI summary
The present invention provides a facial recognition method, including: obtaining a target facial image; determining a covered region and a non-covered region of the target facial image; calculating the weight of the covered region, and calculating the weight of the non-covered region; extracting feature vectors of the covered region, and extracting feature vectors of the non-covered region; comparing the target facial image with each template facial image in a facial database according to the feature vectors of the covered region, the feature vectors of the non-covered region, the weight of the covered region, and the weight of the non-covered region, to calculate a facial similarity between each template facial image and the target facial image; and determining, when at least one of the facial similarities between the template facial images and the target facial image is greater than or equal to a similarity threshold, that facial recognition succeeds.