Real-Time Age and Gender Estimation from Face Images

In this paper, we describe an automated real-time system that estimates age and gender by utilizing a set of facial image sequences from a video camera. The age and gender estimation system consists of four steps: i) detection and extraction of the facial region from input video; ii) selection of the frontal face images from the extracted facial regions using head pose estimation; iii) duplicated face detection and removal by tracking the faces; and iv) age and gender estimation using statistical facial features. Here, LBP features with AdaBoost classifiers are used to detect the face region in a video frame, and the frontal face images are selected using a 3D pose estimation method. In addition, a particle filter-based tracking framework is employed to remove duplicated faces and to improve the accuracy of people counting, and Gabor-LBP features are used to estimate age and gender using a linear SVM and Adaboost classifiers. In experiments, a large number of face datasets are used to train and evaluate the proposed method, and higher performance is achieved in terms of age and gender estimation: 72.53% for age and 98.90% for gender.