Facial Recognition Masking for Obstacle Coverage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition technologies face challenges in accurately identifying faces when they are partially or fully covered by obstacles such as masks, hats, or other objects, leading to reduced accuracy and reliability.
Innovation Solution
A method for facial recognition that focuses on detecting and using the face portion that remains uncovered, generating a mask for the uncovered area, and comparing feature values from this masked region with registered images to determine identity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional facial recognition methods are used to extract features from the entire face area, then recognition speed is maintained, but recognition accuracy deteriorates when the face is covered by obstacles such as masks or hats
Solution Approach 1:
The patent divides the face area into multiple regions (e.g., upper face, lower face, eye region, mouth region) and processes each region separately. This segmentation allows the system to identify and exclude covered regions while maintaining processing of visible regions, thereby improving recognition accuracy without requiring complete reprocessing of the entire face image.
Solution Approach 2:
The patent extracts and removes the obstacle-covered portions from the face image by detecting the obstacle boundaries and creating a masked version of the original image. This extraction isolates the visible face regions, allowing feature extraction to focus only on unrecovered areas, thus improving accuracy while managing computational complexity.
2Measurement precision
If the system attempts to accurately detect and exclude all covered portions using traditional deep learning methods, then recognition accuracy improves, but processing time and computational resources increase significantly
Solution Approach 1:
The patent performs preliminary detection of obstacle presence and basic face landmark identification before detailed feature extraction. By detecting obvious covered regions early in the processing pipeline and applying simple masking operations, the system avoids time-consuming detailed analysis of obviously covered areas, thus reducing overall processing time while maintaining detection accuracy.
Solution Approach 2:
The patent applies partial processing by focusing computational resources only on visible, uncovered face regions rather than processing the entire face image. This partial action approach extracts features only from relevant areas, reducing computational load and processing time while maintaining sufficient detection accuracy for recognition purposes.
3Adaptability or versatility
If feature extraction is performed on the entire face image including hair areas, then processing is simpler, but recognition robustness deteriorates when hairstyle changes occur
Solution Approach 1:
The patent applies different processing strategies to different face regions based on their characteristics. Specifically, it identifies and excludes hair regions from feature extraction by analyzing local image properties and spatial relationships with detected face landmarks. This local quality approach ensures that features are extracted only from stable, non-hair regions, improving robustness to hairstyle changes without requiring complete redesign of the feature extraction pipeline.
Data Source
AI summary
Disclosed herein are a facial recognition method and apparatus based on masking. The facial recognition method based on masking includes performing detection and normalization of a face area including five landmarks in an input image, separating a face portion other than a portion covered by an obstacle in the detected face area, generating a mask corresponding to the separated face portion when there is a portion covered by an obstacle in the separated face portion, applying the mask to an image in a face image registration database and extracting a feature of a masking region, and determining whether it is an identical person based on the extracted feature value.