FACIAL DETECTION IN COMPUTER VISION: BRIDGING GAP BETWEEN CNN, HAAR CASCADE AND MTCNN

Abstract

Face detection is a crucial task in various applications, including face recognition, facial expression analysis, face tracking, and head-pose estimation, spanning fields such as transport, health, and education. Conventional face detectors, from Viola-Jones to CNN-based methods, face challenges in handling diverse facial characteristics in "in the wild" scenarios due to the surge in image and video data. The advent of deep learning brings advancements in face detection, albeit with increased computational demands. This paper provides an overview of deep learning-based methods, offering a comprehensive analysis of their accuracy and efficiency. It delves into a comparative discussion of challenging datasets and associated evaluation metrics. A detailed examination of the efficiency of successful deep learning-based face detectors, including CNN, MTCNN, and Haar cascade, is conducted. The insights gained from this analysis guide the selection of appropriate face detectors for diverse applications and lay the foundation for developing more efficient and accurate detectors in the future.