Face Recognition In Unconstrained Environment

Face recognition has been a challenging research problem to date due to its applications. Researchers are currently focusing on a rare number of images, various face poses, and illuminations. This paper proposes a new promising approach to cope with the face recognition in unconstrained environment. It uses a single face image in various poses (non-frontal face) and various illuminations. It applied Histogram Equalization (HE) to get rid of illuminations and using mirrored images to augment the number of face images. In addition, to cope with non-frontal face with various poses, the face model using Active Appearance Model (AAM) is implemented. The result of AAM is then oriented to the frontal face for feature extraction using Histograms of Oriented Gradients (HOG). For classification, Support Vector Machines (SVM) is implemented. In simulation on FERET benchmark database, the proposed approach provided the outstanding results at 90.12% in accuracy on average superior to the compared method.