Multi-view Face Gender Classification And Feature Selection

This paper studies multi-view face gender classification and feature selection. Gender classification of frontal human face image is already a well-studied topic, but the existing algorithms meet problems at realistic application due to variation of face pose. This paper tries to explore a more adaptive gender classification system of human face image that is robust to pose change, and do feature selection to improve performance. In this paper, the author presents a multi-view face gender classification framework based on face image, and selects feature using HOG analysis and SVM. Experiments are then discussed, and conclusions are made.Face pose estimation has been proved to be a simple task under certain situation. And there has been plenty of work done regarding gender classification study of fixed pose face image. In this paper we combine these two kinds of research. First a face image s pose is estimated, and then the image is classified using the gender classifier which is specially trained with this pose.Gabor-based features have been widely used in face analysis. However, the existing methods usually suffer from high computational complexity of Gabor wavelet transform (GWT), and the Gabor parameters are fixed to a few conventional values which are assumed to be the best choice. In this paper we show that, for some facial analysis applications, the conventional GWT could be simplified by selecting the most discriminating Gabor orientations.Support Vector Machine (SVM) has been widely used in various applications. And a great amount of research has focused on feature selection methods based on SVM. In this paper we try to adopt SVM-RFE and its multi-label generalization, SVM-DFS, to do feature selection on multi-view face gender classification, so that we can accelerate feature extraction and classification process.By analyzing pose classification results, this paper also presents a customized training data set enhancement method, which deals with gender classification accuracy dropping problem when the pose is misclassified. Experiments show that this method improves the overall classification performance.