International Journal of Electrical and Computer Engineering (IJECE) Vol. 10, No. 2, April 2020, pp. 1833~1841 ISSN: 2088-8708, DOI: 10.11591/ijece.v10i2.pp1833-1841  1833 Journal homepage: http://ijece.iaescore.com/index.php/IJECE Optimization of deep learning features for age-invariant face recognition Amal A. Moustafa, Ahmed Elnakib, Nihal F. F. Areed Electronics and Communications Engineering Department, Faculty of Engineering, Mansoura University, Mansoura, Egypt Article Info ABSTRACT Article history: Received Oct 1, 2019 Revised Oct 18, 2019 Accepted Oct 31, 2019 This paper presents a methodology for Age-Invariant Face Recognition (AIFR), based on the optimization of deep learning features. The proposed method extracts deep learning features using transfer deep learning, extracted from the unprocessed face images. To optimize the extracted features, a Genetic Algorithm (GA) procedure is designed in order to select the most relevant features to the problem of identifying a person based on his/her facial images over different ages. For classification, K-Nearest Neighbor (KNN) classifiers with different distance metrics are investigated, i.e., Correlation, Euclidian, Cosine, and Manhattan distance metrics. Experimental results using a Manhattan distance KNN classifier achieve the best Rank-1 recognition rates of 86.2% and 96% on the standard FGNET and MORPH datasets, respectively. Compared to the state-of-the-art methods, our proposed method needs no preprocessing stages. In addition, the experiments show its privilege over other related methods. Keywords: AIFR Deep transfer learning Genetic Algorithm Copyright © 2020 Institute of Advanced Engineering and Science. All rights reserved. Corresponding Author: Ahmed Elnakib, Electronics and Communications Engineering Department, Faculty of Engineering, Mansoura University, El-Gomhoria ST, Mansoura 35516, Dakahlia, Egypt. Email: nakib@mans.edu.eg 1. INTRODUCTION Identification of a person based on his face images over different ages is mandatory for security and forensic applications, e.g., identification of criminals and missing persons. However, this problem is very challenging due to the rapid change in face images with age, especially when images from different age-stages are considered, e.g., newborn, toddler, teenage, and adult face images. In addition, the facial changes are very specific for each person based on his genes, lifestyle. For these reasons, developing a reliable, accurate method for AIFR is necessary In the literature, many research groups have presented different methods to identify person based on facial age images. These methods can be categorized into three categories: (i) generative methods, (ii) discriminative methods, and (iii) deep learning methods. Generative methods construct a personal face model based on the collected facial-age images. For example, shape and intensity features are used by Lanitis et al. [1] to build a 3D age model, which achieves 68.5% recognition accuracy on a private aging data. Pose correction is added, by Park at. al [2], to a 3D shape- and texture-based face model to achieve an accuracy of 37.4% and 79.8% on the two popular online standard databases for AIFR, i.e., FGNET [3] and MORPH-II [4] databases, respectively. The main drawback of these methods is the need for unrealistic parametric assumptions [5]. On the other hand, discriminative methods do not rely on face modeling. They extract direct facial features for classification, such as Gradient Orientation Pyramid (GOP), Scale-Invariant Feature Transform (SIFT) [6], and Multi-Scale Local Binary Pattern (MLBP) [7, 8]. For example, Ling et al. [9] used GOP