Real-Time Gender Classification by Face

Abstract

The identification of human beings based on their biometric body parts, such as face, fingerprint, gait, iris, and voice, plays an important role in electronic applications and has become a popular area of research in image processing. It is also one of the most successful applications of computer–human interaction and understanding. Out of all the abovementioned body parts,the face is one of most popular traits because of its unique features.In fact, individuals can process a face in a variety of ways to classify it by its identity, along with a number of other characteristics, such as gender, ethnicity, and age. Specifically, recognizing human gender is important because people respond differently according to gender. In this paper, we present a robust method that uses global geometry-based features to classify gender and identify age and human beings from video sequences. The features are extracted based on face detection using skin color segmentation and the computed geometric features of the face ellipse region. These geometric features are then used to form the face vector trajectories, which are inputted to a time delay neural network and are trained using the Broyden–Fletcher–Goldfarb–Shanno (BFGS) function. Results show that using the suggested method with our own dataset under an unconstrained condition achieves a 100% classification rate in the training set for all application, as well as 91.2% for gender classification, 88% for age identification, and 83% for human identification in the testing set. In addition, the proposed method establishes the real-time system to be used in three applications with a simple computation for feature extraction.