Novel Method of Face Recognition from Various Pose

Abstract

Camera surveillance and face recognition are useful tools to identify criminal activity and avoid crime. This problem is part of the grand challenge of face recognition variation in pose and camera calibration. Hence, we propose a novel method of face recognition from various pose; the work is based on a geometric method of different pose face that is efficient in handling the range of pose variations within ±60° of rotation, called Face Camera measurement Technique (FCMT). Our approach is to calibrate the camera from a 2D image using fixed and robust facial landmarks to ensure a reliable estimation of real dimensions. Therefore, we decompose the image into grayscale before any feature extraction and selection. Our technique then transforms the original image to a full-face pose, in order to accurately estimate the distance between the eyes. Finally, recognition can be made more stable against a pose difference when we compare the distance between the eyes of the frontal picture as a gallery compared to all the poses. Extensive and systematic experimentation on FERET database shows that our proposed method consistently outperforms single-task-based baselines as well as state-of-the-art methods for the pose problem.