Generate Haar Feature for Face Detection without Mask and Face with Mask

Abstract

The COVID-19 pandemic is a problem that worries the wider community. To stop the spread of COVID-19, the mandatory health protocol to use masks is enforced. Many people do not comply with these health protocols. Based on these problems, a technology was developed to monitor the face that uses a mask or not used. This technology uses the Viola-Jones algorithm. In carrying out its detection function, this algorithm requires a classifier which is the result of training on some positive and negative image data sets. In this study, two positive image data sets were used: facial data using masks and facial data not using masks. The classifier obtained from the training process is a cascade file in XML format that will be used in the detection program. In this study, several training processes were carried out to obtain a good comparison value between positive and negative dataset samples in forming a cascade. The cascade test with the highest accuracy value was obtained from the classifier using 1000 positive samples and 2000 negative samples, namely 98.70% for face detection without a mask and 92.63% for face detection using a mask.