Face Detection in Still Image using SSD MobileNet V2 and Geometrical Algorithm

Face detection plays an important role in today’s technological advancement as it is used in various applications in multiple fields. Many techniques and frameworks have been proposed throughout the decades, utilising different methods that range from algorithms to machine learning and even deep learning. In this paper, a face detection framework that utilises a two-step method was proposed to overcome over-detection in still images containing face images and misdetection in non-face images. The two-step method consists of feature detection using SSD MobileNet V2 and a geometrical algorithm for detecting the face region in the given image. The proposed method was compared with the state-of-the-art face detection algorithms such as MTCNN, Dlib face detector, and OpenCV optimised Haar Classifier. The proposed method performed well in the testing dataset, with 91.5% prediction accuracy. However, it fell short in terms of prediction accuracy performance against the MTCNN and Dlib but performed better in reducing the number of over-detection and misdetection. Hence, achieving the target of the proposed framework.