Gender classification in face images based on stacked-autoencoders method

A gender classification system uses human face from a given image to tell the gender of the given person. An effective gender classification approach is able to promote the improvement of many other applications, including image/video retrieval, security monitor, human-computer interaction, etc. In this paper, a method for gender classification task in frontal face images based on stacked-autoencoders is proposed. Firstly, gender features are learned from frontal face images, followed by dimensionality reduction with stacked-autoencoders algorithm with fine-tuning strategy, which serves as the feature vectors of our method. Ultimately, two kinds of classifiers, SVM and Softmax regression, are trained to the task of classification. The experiment on FERET and CAS-PEAL-R1 face datasets is reported that an effective method is proposed for gender classification task and other methods are compared with ours.