Age And Gender Estimation By Using Facial Images

Face images contain abundant of information, such as gender, age, race, expression, etc. The application based on face image is now widely attention, and face recognition system has been put into use, research on face used also goes deeper. The gender classification and age estimation based on face images has become a hot topic, such as the ratio of female and male in supermarket and age distribution. In order to raise the recognition accuracy, this paper does some further study on gender recognition and age estimation,and suggests a method for estimation.The main contents are as follows:(1) Face normalization. Images obtained from different databases have so me differences, such as different sizes, face areas, rotation, etc. So it is necessary to do some preprocessing for the follow operations. This paper first use Adaboost algorithm and Haar feature to automatic detect the face from the image, and then normalize the face image based on eyes and mouth. Thus, reduce the impact on location information and background noise.(2) Face features extraction. The pixel-value as the global feature and Local Binary Patten as the local feature are used for gender recognition. Geometric and texture features are used in age estimation. Geometric features are extracted by using active shape model(ASM) and texture features are obtained through Gabor wavelets transform and fractional differential.(3) Face features dimension reduction. Gabor wavelets texture feature extraction will rapidly intensified features’ dimensions, this may consume a lot of time and the redundant information may decline the recognition accuracy. This paper uses principal component analysis(PCA) which is the good for data dimensionality reduction to reduce the features’ dimensions, and compares and analysis the accuracy.(4) Face gender recognition. Gender recognition is a two class classification problem, this paper uses support vector machine training and testing the gender characteristics.(5) Face age estimation. Age estimation is a multi-classification problem, in this paper, integrating support vector machine classification and regression to solve the problem. Firstly using support vector machine method to recognize children and adult, then using SVM regression to train and test the adults by using texture features, and results obtained through the weighted method; Secondly, to classify the results to the age range by SVM; Finally, combine the geometric features and texture features to classify children, adults, the elderly.This paper proves that in the classification of the children and adults through the fusion of the geometric features and the distance of the features’ points, the average accuracy rate can be increased by about 3%. Moreover, the combination of different texture features can improve the accuracy of gender classification and age estimation.