Detecting Face Masks with Machine Learning
Masks are intended to save the wearer from spreading viral droplets and other respiratory infections. Take a look at the whitepaper to learn how a machine learning-based model will help detect face masks.
The whitepaper explains the execution of a facial recognition system using a global method to feature extraction based on a histogram-oriented gradient. Feature vectors for multiple faces were extracted from the Yale and AT&T databases and used to train a binary-tree structure SVM learning model. Running the model on both databases resulted in over 90% accuracy in matching the input face to the valid person from the gallery. We also documented one of the drawbacks of using a global approach to feature extraction of a model trained using a feature vector of the entire face in place of its geometrical makes it less potent to orientation and angle changes. Nevertheless, when the deviation in facial orientation is not substantial, the global approach is still exact and more specific to execute than component-based approaches.
This paper also proposes the MobileNet-based Depthwise Separable Convolution Neural Network (DS-CNN) for mask detection in facial images. We evaluate our findings on the unique convolutional filters on specific datasets.