Gender Studies Based On The Feather Extraction Of Face Images Analysis

The main object of study is the gender recognition based on face image. Detect and locate the human face images, determine whether there are images of face, how much faces there are, and the positions of the face images. Second, preprocess the face image to eliminate the impact of changing light and noise, and ensure to get better recognition results. The image preprocessing includes gray image, geometry and energy normalized and histogram equalization. The facial feature extraction is based on the first two steps of processing the images. the feature extraction is to extract the facial features affect the gender classification, including beard feature, hair length feature, hair thickness feature, eyebrow thickness feature, which are the key features to identify the gender. The BP neural network is trained, the input of BP neural network is composed of the four features of a column vector, and the output is gender information of the human face images.The face gender recognition of face images is based on the detection and localization of face images. The face detection algorithm is implemented by the adaptive AdaBoost algorithm and the PCA algorithm. The adaptive AdaBoost algorithm is achieved by changing the distribution of data to exclude the non-face image area, so achieves the accurate detection and location of face region. The PCA algorithm uses the"feature-face" approach to achieve the purpose of face detection, which determines whether it belongs to the face according to the "face space" the distance of the sample. The focus of face gender is the face image feature extraction, and the key of feature extraction is the correct location of the eyes. The eye location method is presented based on the corner of information of eyes to improve the right rate of detection. Obtain the facial feature as the BP neural network, such as characteristics of the beard, hair length and thickness characteristics, and eyebrow thickness characteristics. BP neural network is trained to adjust the weights to satisfied error requirement with self-built face database.Experimental results show that: The correctly detected is 82%. For most face images, the right gender information are given. There are a small number of incorrect results, and the reason is that there are no significant gender characteristics of the face images or the gender characteristics of face images appears in'Reverse'way. The method has higher detection accuracy to achieve the desired requirements.