Gender classification using selected independent-features based on Genetic Algorithm

The Gender of a face is almost its most salient feature, and realizing automatic gender classification according to the face image will boost the performance of face retrieval and face recognition in large face database. This paper proposed a new gender classification method combining independent component features selected based on genetic algorithm and support vector machine (SVM). First, the FastICA algorithm is used to derive independent basis image vectors out of the training face images. Each image is represented as a feature vector projected in the low-dimensional space spanned by the basis vectors. Then, A Genetic Algorithm is used to select a subset of features which seem to encode important information about gender from the low-dimensional representation. Finally, the SVM classifier is trained to perform gender classification using the selected independent-features subset. The experiment results show that the method gets a better classifier performance.