Masked Face Recognition via Synthetic Data Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Facial recognition systems struggle to recognize masked faces due to the lack of representative samples in existing training datasets, leading to ineffective occlusion handling.
Innovation Solution
The method involves generating an augmented training dataset by adding images of masks to original face images using affine transformations and generative adversarial networks (GANs), ensuring the neural network model can recognize masked faces by simulating various occlusion scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If facial recognition systems use existing training datasets with clear face images, then the training process is simple and fast, but the system fails to recognize masked faces accurately
Solution Approach 1:
The system performs preliminary data augmentation by generating synthetic masked face images before training the neural network. Mask images are created and superimposed on clear face images from existing datasets, preparing the training data in advance to teach the system how to recognize masked faces without requiring actual masked face photographs.
Solution Approach 2:
The system creates copies of existing clear face images and modifies them by superimposing mask images. These synthetic copies serve as training examples, allowing the neural network to learn masked face recognition patterns without needing real masked face data, thus resolving the shortage of representative samples.
2Reliability
If manual curation of masked face data is performed, then the training data quality improves, but the time and resources required increase significantly
Solution Approach 1:
The system automatically generates its own training data by programmatically creating mask images and superimposing them on existing face images. This self-service approach eliminates the need for manual collection, annotation, and curation of masked face photographs, significantly reducing time and human resources while maintaining high training data quality.
Solution Approach 2:
The system introduces an intermediary synthetic data generation process between the existing clear face images and the neural network training. Instead of manually curating real masked photos, the system uses automated mask superimposition as an intermediary step to create realistic training examples, saving considerable time and resources.
3Adaptability or versatility
If the training dataset lacks diverse occlusion scenarios, then the data processing is simpler, but the system's adaptability to different mask types deteriorates
Solution Approach 1:
The system varies parameters of the synthetic mask generation process, including different mask positions, opacities, sizes, and superimposition methods. By changing these parameters programmatically, the system generates diverse occlusion scenarios covering various mask types without manually creating separate datasets for each variation, thus improving adaptability while managing complexity.
Data Source
AI summary
Methods and systems for training a neural network include generate an image of a mask. A copy of an image is generated from an original set of training data. The copy is altered to add the image of a mask to a face detected within the copy. An augmented set of training data is generated that includes the original set of training data and the altered copy. A neural network model is trained to recognize masked faces using the augmented set of training data.