Research On The Technology Of Age Estimation Based On Gender Identification

Human face recognition is attractive in pattern recognition and image processing. It has been applied to security surveillance, access control, artificial intelligence, HCI and so on. However, the recognition rate will decline sharply due to the change of shape and texture in the face for different ages and genders. Besides, the variations of different gender images are different with the changing of age. To solve this problem, we considered both gender identification and age estimation, and bring them into face recognition algorithm which based on the present face recognition and age change research. Thereby we present the age estimation which allows gender change. We designed and implemented the prototype system of age estimation.The highlights and main contributions of the dissertation include:(1) Combination of Gabor wavelet and KLDA is used to extract feature. After analysing the advantages and disadvantages of present method of feature extraction, two-dimensional Gabor wavelet is used to extract the key characteristics in gender identification, and then use KLDA to extract the features which are extracted by Gabor again to obtain the final feature vector, which has the biggest between-classes scatter and smallest within-class scatter. KLDA can be applied to reduce the extracted feature inunlinear space. Then, these features are helpful in gender identification.The experimental results have shown that the approach can improve the recognition rate.(2) An improvement algorithm which based on combination of SVM and FKNN is presented to improve the accuracy of gender recognition nearby SVM hyperplane. FKNN combied SVM can weaken the impact of the disparity of training samples in distribution on categorization performance with different feature selection based on the analysis of image similarity, membership and distance. The algorithm gets optimal threshold by a few of known gender samples, then compute the distances from the test samples to the optimal superplane of SVM in feathure space, recognize gender after comparing the distance to threshold. Therefore, this algorithm can improve the accuracy of classification, reducing the sensitivity of the value of k.(3) An age prediction algorithm based on gender identification. This paper will present the method of ASM-AAM to build a face model,then using this model abstract shape feature and texture feature. This method either makes use of the advantage of ASM and AAM or avoids the disadvantage of ASM and AAM. ASM is good at to locate the shape model but it cannot locate the texture model. AAM is good at to locate the texture model but it cannot locate the shape model accurately. After building the face model, a two-model of male and female is used to exstract the age feature.(4) Based on the idea of oriented object, we design and implemt prototype system of age estimatiom with gender variances, which is divided into four functional modules that is image preprocess, feature extraction, gender identification and age estimation. The results of experiment prove that the algorithms above are effective in age estimation in training images and test images of different gender.