Study On Key Technology In Gender Classification And Age Estimation Of Human Face

In recent years, face recognition technology has become a hot topic in the field of pattern recognition, and it attempts to give the computer the ability to distinguish their identity based on the people’s faces, the study has important theoretical significance and broad application prospects. The human face is one of humanity’s most important biometric, it contains a lot of important information, such as identity, gender, age, expression, race and so on. The demographic characteristics of gender and age have important reference significance for identity discrimination, and there are potential applications in human-computer interaction, computer vision and business intelligence. But under uncontrolled conditions, however, due to various expressions, illumination, poses as well as shooting conditions and they greatly affect the recognition rate of the algorithm, so real-time gender classification and age estimation on human face is a very challenging task.1. This paper will put gender classification and age estimation of two aspects for discussion and research respectively, which mainly include:the face image preprocessing (image normalized and histogram equalization), biological feature extraction, feature selection, feature dimension reduction, manifold learning, pattern classification and regression, then through integrate weak classifiers to form a robust strong classifier for gender problem to complete this classification task and do local adjustment based on the regression results to improve estimate accuracy for age problem. Finally, we finished an automatic gender discrimination and age estimation demo system on human face based on the theoretical study of this paper. The main research works are summarized as follows:2. Proposed an efficient gender classification method on human face, this method integrated simplified Gabor feature extraction, multi-clustering feature selection (MCFS), side information linear discriminant analysis (SILD) and support vector machine (SVM) as a whole. Which MCFS is used for local feature selection, SILD is used for local feature dimensionality reduction and SVM is used for weak classify. Finally, several weak classifiers will be combined into a robust strong classifier to complete gender classification.3. Proposed the concept of age label relative relationship, and embedded the concept into relevant component analysis (RCA) and orthogonal locality preserving projection (OLPP) to form a more discriminant manifold data. After obtained robust feature manifold subspace, in order to predict the age more accurately, we have designed a local adjustment based on the regression results and take full advantage of the partial training data to improve the accuracy of age estimation algorithm, named local adjustment of relative attributes (LARA) to complete the final age prediction.Achieved an automatic face gender classification and age estimation demo system. Gender classification and age estimation both shared in face detection, the feature points location and image preprocessing three big modules, and the left gender classification module is divided into simplified Gabor feature extraction, multi-clustering feature selection (MCFS), side information linear discriminant analysis (SILD), weak classifiers integrated four small modules and age estimation module is divided into Gabor feature extraction, label relative of relevant component analysis (lrRCA), label relative of orthogonal locality preserving projection (lrOLPP), support vector regression (SVR), local adjustment of relative attributes (LARA) five small modules.