Research On Multiple Scale Textural Feature Based Gender Recognition

Gender recognition technology has been widely used in security monitoring, authentication, video retrieval, robot vision, and human-machine interface, and has drawn wide concern in the areas of pattern recognition and computer vision, however, how to let the computer acquire the same capacity in gender recognition as human being remains a challenging research topic.Gender recognition system includes three main parts, including the image data preprocessing, feature extraction, as well as training the classifier. In the process of image acquisition, due to the environmental factors (especially light and background), wearing ornaments as well as body posture, the image data captured containing different levels of noise, we need to preprocess it to improve the accuracy of the system by reducing the impact of the image itself.Generally, in gender recognition technologies, features such as HOG feature, Haar-Like feature and textural feature are widely used. In this paper, we studied comprehensive comparison of these characteristics and proposed the multiple scale textural feature used in our algorithm. Combined with the improved AdaBoost algorithm, we tested our gender recognition algorithm on the basis of MIT database and simulated it on the IOS system. The experimental results used in this paper show that, with the same eigenvalue, the improved gender classification algorithm acquire higher efficiency and accuracy, and with the same classification algorithm, the multiple scale textural feature tends to be more robust. Therefore, the proposed gender classification algorithm exhibit higher performance either in terms of the identification efficiency or the recognition accuracy and is promising in the applications of mobile computing.