A Study Of Gender Classification Based On Face Parts And The Complementarity Of Face Parts

In recent years, Biometrics Research which based on the face images has made great development . Compared with other biometrics, for its advantages, such as natural, convenient, and non-contact, face feature has great prospect in security monitoring, authentication, human-computer interaction, and so . Gender recognition which based on face images means the issue of pattern recognition according to face images to recognize genders. Gender recognition becomes one of the hottest research topics in the field of computer vision and pattern recognition, and also got much attention, because of its potential applications in identity authentication, human-machine interface, video retrieval and robot vision.Integrated access and analysis of the current research of face gender recognition, it found that most gender recognition issue are to extract feature of full face image for recognition. However, this feature recognition is not good for the face images which have the existence of occlusion or pollution of to its relatively poor robustness, while full face recognition are also vulnerable to changes of facial expressions. In contrast, the facial expression changes have little effect on sub-region, for some partial invisible face image can recognize only by its visible part. gender identification method of sub-regional and their combinations proposed in this paper.In this paper, the writer clearly introduced the contributions of every part of face, eyes, nose, mouth, chin, and the full face field(eyes, nose, mouth and chin included). Select a image that comes from the FERET image database, preprocessing it, standardize it, extract five sub-regions. The experiment extract the efficient features by PCA which decrease the dimensions of the picture, and then classify the image by LIBSVM, and the results show that the five sub-regions have enough gender information and the right ratio of gender recognition reaches 80%.As every sub-region of face contains complimentary gender information, this paper show the experiment which takes advantage of different of combinations of five sub-region based on single sub-region recognition. The experiment uses the SVM as base-classifier, simple voting and weighted-voting methods as integrated classifier to realize the combination recognition of five sub-regions. The experiment primarily researched the recognition of the combination of 3 sub-regions, and the result show the it is as good as 85%, and this method is better when the image is polluted or partial occlusions...