Facial Features Extraction for Age Estimation Based on a Neural Network

Abstract

The determination of a person’s age using computer-based programming relies on biometric features such as facial features and fingerprints. These technologies, which require the intense study of age progression mechanisms in the human face and the factors that influence face aging patterns, have increasingly attracted attention due to their various real-world applications. In this paper, a feature selection method (FSM) has been adopted to identify the most suitable local binary pattern (LBP) features for age estimation. The selected features were then classified using a back-propagation neural network (BPNN). Extracted features were classified using a MATLAB-based artificial neural network (ANN), in which a back-propagation network (BPN) was used. The proposed BPN network contains four layers of neurons: an input layer, two hidden layers, and an output layer. Each layer contains nine neurons in addition to the bias. Experiments in this work were conducted using three types of image datasets: standard FG-NET, privately-collected, and real-time captured image dataset. Experimental results demonstrate the significant role of using a back-propagation neural network (BPNN) with FSM, which produced an estimation accuracy of up to 93.22% after two rounds, while without FMS, estimation accuracy reached 82.966% at most, depending on the original LBP features. However, the accuracy using the images captured by the connected camera reached no more than 71%.