Research Of Gender Classification Based On Facial Images

In this paper, the main content is a problem of gender classification based on facial images. Face is one of the most important biological features, which reflects a lot of biological information on identity, gender, expression, ages, ethnicity, etc. With the rapid development of computer technology, computer vision and pattern recognition based on facial images have also become hot issues in recent years. Most researches based on facial images focus on identity recognition, which is only to distinguish who is the input image. As an important component of face recognition, the gender classification based on facial images not only helps provide a more personalized human-computer interaction, but also can be applied to information collection system and video surveillance system. From the theoretical point of view, the research on gender classification of facial images also enriches the existing methods; improves the accuracy of face recognition and the efficiency of image retrieval. Therefore, the research on gender classification based on facial images is significant.In general, a system of gender classification based on facial images consists of image preprocessing, feature extraction and classifier recognition. In this paper, the research and innovation are executed in these three aspects.1) In the preprocessing, a color image generally should be converted into grayscale, with using histogram equalization method to enhance the image. Then the geometry normalization based on eye location and the energy normalization should be done. In this paper, a Gabor method is used in the eye location.2) In the research of face recognition, many effective feature extraction algorithms are widely available. This paper studies the PCA, LBP, PCA-SIFT algorithms, and in accordance with the characteristics of gender classification, an Enhanced PCA-SIFT algorithm is proposed to make the feature vectors contains initial gender information and discrimination, thus facilitating the following classifier's work.3) In the step of classifier research and design, two representative algorithms, SVM and FSVM are researched in depth. The thought of fuzzy is introduced into the gender classification, and a membership function designed by the Enhanced PCA-SIFT is proposed, which improves the generalization ability of FSVM. In this dissertation, we choose FERET, CAS-PEAL and BUAA-IRIP and the laboratory self-built facial image sets as the experimental facial image databases. Throughout the contrast experiments, the results show that Enhanced PCA-SIFT can extract feature vectors which have initial gender characteristics, and FSVM proves a good ability of classification.