Gender Recognition Based on Face Image

Abstract

Biometric systems are usually associated with the use of unique physiological characteristics such as fingerprints, hand geometry, retina, iris, and hand signatures to identify an individual. The human face is considered as one of the most important biometric traits that contains information about the subject gender, race, age, and mood. Face-based person recognition/identification is challenged by many problems including the detection of the face region and its landmarks. However, current deployments of large scale face based biometric systems suffer from large processing times due to the steadily growing face databases used by these systems. Any reduction in processing times has great impact on the overall system performance. In this thesis, gender recognition is proposed to reduce the search space of face-based person identification/recognition systems. Other factors may be considered as well such as skin color and face expression. The objective of this thesis is to develop a face-based gender recognition algorithm using various image features. Statistical features are given special attention for their ability to represent better the face landmarks. In this thesis, local face regions are represented using the GIST, pyramid histogram of oriented gradients (PHOG), GIST based on the discrete cosine transform (DCT) and the principal component analysis (PCA) features. These features are extracted using the face local regions. Then, gender classification is carried out using a support vector machine (SVM) classifier on these features. Finally, the performance of the proposed features and classifier is evaluated against state-of-the-art gender classification techniques using face images acquired in uncontrolled environments.