A Deep Cascaded Multi-task Face Recognition Framework

Abstract

In an unhindered environment, the detection of face and alignment are more challenging due to a variety of poses, illuminations, and occlusions. To better understand and boost, a deep cascaded multi-task framework is proposed in this paper, which exploits alignment and detection inherent correlation. In a coarse and fine prediction of landmark location of the face, the proposed framework leverages Multi-task Cascaded Convolution Neural network [1] (MTCNN) followed by FaceNet [2] to recognize the face identities efficiently. The work has been extended in the form of a real-time attendance marking system. The proposed technique achieves better recognition performance compared to the other state-of-the-arts. Three publicly available datasets: ORL, AR, LFW, datasets are used for experimentation.