Gender Classification by Using Edge Oriented Histograms and Edge Density of Face Images and AdaBoost Algorithm

• Main Authors: Ta-Li Liu, 劉大立
• Other Authors: Zhi-Fang Yang
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/29629740970233980252

碩士 === 國立臺北大學 === 資訊工程學系 === 101 === Gender has always been an important trait of human being. It is the most direct way to use face features to identify one’s gender. With the advance of computer vision of recognition technology, many good recognition techniques have been developed in the past days. However, it is still a challenge to identify the gender from one’s face image by catering to human vision and think in an intuitive way. Therefore, how to develop a more effectively and directly identify method, and also can meet practical business and security use purposes for a more convenient and secure living environment is an important challenge nowadays. The subject of this thesis is to discuss the achievements of gender recognition from face image after edge detection. The central idea primarily based on face gender identification by human intuitive vision. Since facial features and contour on female’s face are delicate and more curves, but on male’s face are rugged and little curves on male's face, the gender identification from human face can be achieved through a training method based on learning theory. In this study, we propose the edge direction histogram (EOH) method to capture the characteristics of two genders, and then use the edge density (ED) to capture another feature. We chose the AdaBoost learning algorithm to classify the various cached features to weakclassifier, and produce a strong classifier from this training algorithm. In the experiment, we pre-processed the face images in the database, and then though the retrieving, training and testing process. A large window was used to extract features for reducing the massive training time, . The results show that some changes of the magnitude threshold can improve the recognition accuracy, and also confirm that female has more directionality in some aspects than male. The cross-validation method was applied on the whole training and testing to prove that after edge direction on the face image, and then use edge direction histogram (EOH) and edge density (ED) features is feasible for gender recognition and compliance with human vision. A new system for gender recognition on face is hence successfully proposed in this research.