Evaluation of a DNN Face Recognition System Based on Aging Characterization of a Twin Set

Abstract

Development of deep learning techniques plays an important role in the successful extraction of distinctive features that represent objects in images. This development has contributed significantly to obtaining unique features from individuals' facial photographs with deep learning methods. A face photograph is passed through the convolutional neural networks (CNN), and feature vectors of a certain length are produced that represent unique features for each person. The feature vectors produced differ according to the structure of the deep learning algorithm. The facial recognition model used in this study was developed by training about 3 million images, and it produces 128-length, unique vectors for each person. These vectors are classified with k-nn classifier and face recognition is made according to the appropriate k parameter selected. In the study, the method applied for solving the aging and twinning situations with deep learning, which is one of the important problems encountered in face recognition, is discussed. With deep learning, face recognition was made using the prepared twins' dataset, and Euclidean distance analysis of the produced feature vectors was done for twins. The model was trained using old photographs of the people and k-nn classifier, and the current photographs and results of the people were tested.