Human identity recognition based on facial images: Via supervised autoencoder features representation
Saddam M. Eragi, Wael Ouarda, Adel M. Alimi · 2017
Face recognition is still a challenging field due to the wide range of its applications like security, surveillance, criminal justice systems, witness face reconstruction etc. Recently, Researchers achieve an excellent performance on this task by using deep learning. In this paper, we propose a novel approach of features representation of facial images using Supervised Autoencoder. The main goal of this paper is the representation of facial images by a combination of descriptors, such as Viola and Jones technique as a face detector, LBP, HOG, Gabor, Curvelet and Wavelet for features representation, Stacked Autoencoder to transform features and finally Linear SVM as a classifier. The experimental results on Japanese Female Facial Expression Database (JAFFE) and Cohn-Kanade database (CK) have shown the robustness of the Autoencoder as technique for features representation and transformation in nonlinear space. Experiments performed on these databases highlight the effectiveness of our proposed approach by enhancing the Recognition Rate of state of the art to 98.6% on (JAFFE) database and 99.4% on (CK) database.