COJ Electronics & Communications

Robust Representation for Face Recognition in the Wild

Submission: April 07, 2020; Published: June 02, 2021


This paper describes a generalized representation for face recognition in the wild using still and video imaging, addressing unconstrained recognition under A-PIE effects. The proposed representation uses images and video frames for generating the facial signatures for face recognition. The representation allows for inclusion of all video frames in generating the facial signature and is also amenable to “best frames” selection and “intelligent priors” about facial regions in the image/video. The representation includes dual optimization for signature extraction and signature matching over large gallery manifolds. It is modular and amenable to deployment on distributed systems, smart phones and on the cloud. Results, in a Face Recognition at a Distance (FRAD) setup, for detection of faces under severe occlusion, pose-invariant face detection and recognition, on databases created by high resolution camera (Canon 7D: 18MP) and iPhone 4 low-resolution camera (5MP), give credence to the proposed representation.

Keywords: Face recognition in the wild; Selective Part Model (SPM); Gallery manifolds; Pose-invariance; Illumination models; Deep learning; Feature selection; Matching

Get access to the full text of this article