Weight-Lightened Face Recognition Model for Masked Identity Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current face recognition systems face challenges in accurately identifying masked faces due to reduced recognition speed, high time and cost for database configuration, and potential security issues during real-time dynamic service processes, especially when transmitting face information to external servers.
Innovation Solution
A face identification apparatus and method utilizing a weight-lightened face feature extraction model through techniques like network pruning and quantization, which detects and aligns face features, generates a mask synthesis learning database, and employs a voting-based approach to minimize misrecognition and secure edge environment processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a separate module is used to extract eye parts for mask-wearing face recognition, then recognition capability for masked faces is improved, but recognition speed is lowered
Solution Approach 1:
The patent merges the mask-wearing face recognition capability into the main face recognition model through knowledge distillation, eliminating the need for a separate module. The student model learns to recognize both normal and masked faces simultaneously during training, achieving unified recognition without additional processing steps that would slow down recognition speed.
Solution Approach 2:
The patent segments the training process into two phases: training on normal faces first, then training on masked faces with knowledge distillation. This segmented approach allows the model to build foundational face recognition skills before specializing in masked face recognition, improving both capability and efficiency.
2Measurement precision
If a database of faces wearing masks is constructed for learning and recognition, then recognition accuracy for masked faces is improved, but time and cost consumption increases
Solution Approach 1:
The patent uses knowledge distillation to copy knowledge from a pre-trained face recognition model (teacher model) to a student model specialized for masked faces. This approach avoids the need to construct and train on a large database of masked faces from scratch, significantly reducing time and cost while achieving accurate recognition.
Solution Approach 2:
The patent performs preliminary training on normal faces before training on masked faces. This preliminary action builds a strong foundation in the model, allowing subsequent masked face training to be more efficient and require less data, thereby reducing overall time and cost consumption.
3Measurement precision
If face information is transmitted to an external server for recognition, then recognition capability is improved, but security issues may occur
Solution Approach 1:
The patent extracts the face recognition capability from an external server environment and embeds it directly into the local device through the student model. This extraction eliminates the need to transmit sensitive face information over networks, removing the security vulnerability while preserving recognition capability locally.
4Measurement precision
If the face feature extraction model is optimized for accuracy, then recognition precision is improved, but processing time and computational resources increase
Solution Approach 1:
The patent changes the parameters of the face feature extraction model through knowledge distillation, where a large teacher model's knowledge is transferred to a smaller student model. This parameter transformation allows the student model to achieve high recognition precision with reduced computational complexity and faster processing time.
Data Source
AI summary
The present disclosure relates to a face identification apparatus and a method thereof. An apparatus includes: a processor configured to detect face information from image data, to extract face features from the detected face information based on a face feature extraction model weight-lightened through a weight-lightening technique, and to compare the face features with a previously stored face database to identify a face according to similarity; and a storage configured to store data and algorithms driven by the processor, and to have the face database.