Unconstrained Face Recognition
In the present day, Face Recognition (FR) is one of the most relevant and intensely studied problems in computer vision. The most common way of distinguishing and identifying a person or many persons is through the face. FR is achieved through deploying systems, which recognize one or more persons in the still images or videos. Video Face Recognition (VFR) has emerged to be a potential area of research due to the demands arising in the area of commercial applications, security systems and law enforcement systems. The extensive set of applications include content-based video indexing/search, biometric identification, video surveillance and many other applications made the study on VFR inevitable. The significant intra/inter-class variations due to low quality of the video, occlusion, motion blur, unconstrained acquisition and a large amount of data to be processed made the video-based FR more challenging compared to still image-based FR.
With the increasing availability of GPUs and powerful TPUs for massive data computation, deep convolutional neural networks have proved to be very successful in face recognition from still images and video. The power of Deep Convolutional Neural Networks (DCNN) makes it possible to produce robust deep representations to detect faces more accurately from videos than traditional approaches.
The principal aim is to develop a deep learning based Unconstrained Face Recognition system to address the difficulties arising due to the low quality, pose variations, motion-blur etc. in surveillance video
Gopakumar, Rajesh, "Unconstrained Face Recognition" (2022). Technical Collection. 53.