Research Of Gender Recognition Based On Face Image

Among all the biological characteristics, human face recognition has always been a hotissue for its integration of gender, identity, race, age and other important information.Biometric characteristics identification technology is a kind of process that combinesbehavioral characteristics and physiological features through reliable means of analysis andclassification. It has the main features of the unique, security and stability. Automatic humangender recognition by the computer has become a vital task in the field of computer visionresearch. Face gender recognition is based on face identification.Long term study of faceidentification has greatly promoted the development of the human face gender recognition.Generally, human face based gender recognition involves three parts: human face detectionand image preprocessing, facial characteristics extraction and classifier designing. In thispaper, the recognition performances among different designs have been compared. It makesdetailed research on these three parts. The main work is as follows:(1). The history and present situation of the human face are summarized based ongender recognition. From current references, an overview of new methods is provided, andespecially a classification of different methods of human face is made based on genderrecognition nowadays, including summarizes feature extraction and classification methods inhuman face gender recognition.(2). The human face detection is discussed based on AdaBoost, which means theextraction of human face region from a common image so as to have the characteristicsextracted and classifier designed. The purpose of the face detection is to obtain face regionfrom a natural image. In order to obtain a valid image, pre-processing for face region image isneeded for feature extraction and classifier design.The current methods of human facedetection are summarized and classified. Discussions about the AdaBoost algorithm aremainly made. The results of the detection of a single human face and multi-human faces areshowed. Also the results of the detection of facial sub-domain are showed. Besides, thecommon image preprocessing is introduced and the effects after processing demonstrated.(3). The characteristic point orientation on the basis of Active Shape Model (ASM) isexpounded. The establishment and searching procedures of ASM is briefly introduced. Themanual localization of the characteristic points is modified. The key points of face promoterregion are got by certain rules. The outline of face promoter region is obtained through ASM.The procedures of automatic localization of the characteristic points are completed, and then geometry characteristics on the basis of characteristic points are extracted.(4). The extraction of the characteristics of human facial sub-regions on the basis ofGabor transformation is explored, and the definition of one dimension and two-dimensionGabor transformation are briefly introduced. The principles of Gabor filter are explained, andthe human face characteristics extraction according to local Gabor transformation is mainlyexpounded.(5). The classification of support vector machine (SVM) is expounded. The SVM theoryis based upon the theory of minimization of structural risk, enjoying many unique advantageswhen dealing with dichotomy issues such as gender. With the same training samples andtesting samples, the distinguishing accuracy of three kernel functions are compared andfinally the radial basis function (RBF) is selected to be the kernel function, and LIBSVMalgorithm is taken to train and distinguish the geometry features and Gabor sub-regionfeatures.(6). The classification of geometry features and the sub-regions of the Gabor localtransformation are fused. By using min-max method to have the primary data standardize asthe input data of the classifier. The strategy of decision-level fusion is used, and the simplemajority voting is refined. The weighed majority voting is introduced in order to improve theaccuracy of recognition.