Comparative Study of Face Detection Methods for Robust Face Recognition Systems

Face detection systems are used in various computer vision-based applications such as biometrics, security, surveillance, etc. Computationally immoderate face detection methods may not be convenient for devices with inadequate resources. On the other hand, an appropriate face detection approach should be considered in order to achieve high accuracy and substantial performance. This paper deliberates different methods of facial detection and contrasts them to find a better approach for a robust facial recognition system. Five methods of face detection were used in this comparison namely, ViolaJones, Histogram of Oriented Gradient with Support Vector Machine (HOG-SVM), Multi-task Cascaded Convolutional Network (MTCNN), Single Shot Multibox Detector (SSD) and Maxmargin Object Detection (MMOD). Each method was evaluated by varying illumination intensity, angle of the face, the scale of the face and different occlusion types. Video data and WIDERFACE image samples were used for the analysis. Obtained experimental results depict that SSD performs better on the task of face detection with high accuracy and performance, while MMOD has the lowest performance and Viola-Jones gives the lowest accuracy.