Designer Face Mask Detection Using Marker-Based Watershed Transform and YOLOv2 CNN Model

Face mask detection using artificial intelligence (AI) has become more challenging due to high variability in modern designer masks such as single color, multicolor, textile, and printed. This paper presents a designer face mask detection using marker-controlled watershed transform and YOLOv2 CNN model. In the first stage, the face images are preprocessed and segmented using marker-controlled watershed transform by setting mask color as foreground and face color as background marker. The segmented image is applied to the YOLOv2 CNN model for the detection of the face mask. Marker-controlled watershed transform is employed for segmentation and highlights the multicolor mask texture, to improve the classification efficiency of the YOLOv2 CNN model. Simulation performed using different types of designer masks gives an accuracy of 86.66% and an F1-score of about 0.91, which verifies the efficiency of the proposed scheme. The technique deployed in this paper can be used to develop automated systems for face mask detection, classification, and alarm systems.