Face Recognition Using Eye and Forehead Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional face recognition systems experience significant accuracy degradation when dealing with masked faces, particularly due to the lack of sufficient training data, damage to mouth and nose features, and challenges in detecting faces under mask occlusion.
Innovation Solution
The proposed face recognition method focuses on using the eyes and forehead for recognition, omitting the need for reconstructing occluded areas, and employs a CNN-based feature extractor with preprocessing steps like smoothing, cropping, and grayscale conversion, along with covariance deep feature extraction and Bag-of-Features paradigm for classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional face recognition techniques are used on masked faces, then the system can process face images, but the recognition accuracy degrades significantly (by 20-50%) due to mask occlusion covering 50-80% of the face
Solution Approach 1:
The patent extracts and utilizes only the unoccluded facial regions (eyes, forehead) for recognition, completely ignoring the masked areas. This is achieved through cropping the face image to retain only the upper portion above the mask, then extracting features from this cropped region using deep learning models, thereby eliminating the harmful effect of mask occlusion on recognition accuracy
Solution Approach 2:
The patent segments the face image into occluded and non-occluded regions, then selectively processes only the non-occluded segments for feature extraction and recognition. This segmentation approach allows the system to work with partial facial information while maintaining high accuracy
2Adaptability or versatility
If the system is trained with masked faces data, then it can handle mask occlusion, but the training requires collecting and annotating millions of masked face images which is time and energy consuming
Solution Approach 1:
The patent applies preliminary cropping action to remove masked regions before feature extraction and training. By pre-processing the images to eliminate occluded areas, the system reduces the complexity of training data requirements and accelerates the training process while maintaining adaptability to masked face scenarios
Solution Approach 2:
The patent creates a simplified representation of masked faces by copying only the visible upper facial regions into the training dataset, replacing the need to process full masked face images. This copying approach reduces the amount of training data needed while preserving the essential recognition features
3Measurement precision
If the system uses deep learning-based algorithms for face recognition, then it achieves high accuracy in controlled environments, but it suffers from extreme degradation in uncontrolled environments with illumination variations, pose variations, and occlusion
Solution Approach 1:
The patent converts the harmful effect of mask occlusion into a beneficial filtering mechanism by using the mask itself as a guide to crop and isolate the useful unoccluded facial regions. This transforms the occlusion problem into an advantage for focusing feature extraction on the most informative areas of the face
Data Source
AI summary
The present invention relates to a face recognition method that enables to recognize masked and non-masked faces with high accuracy.