基于特征融合的人脸图像性别识别

Facial image gender recognition method based on feature fusion

在人脸图像的性别识别方法研究中,存在同一个人既参与训练又参与测试的情况,所得结论有一定的局限性.针对此问题,建立相互独立的测试集和训练集.传统性别识别模型,受相关参数影响较大,稳定性有待提高,为此,提出一种基于特征融合的人脸图像性别识别方法,采用主成分分析和正交化的线性判别分析相结合的方法表述图像的全局特征,突破传统线性判别分析二分类时秩的限制,采用均衡的局部二值模式方法表述图像的局部特征,将少量全局特征和局部特征相融合,形成人脸图像的性别特征.支持向量机用于实现性别特征的分类.实验结果表明,此方法在具有一定稳定性的同时,能获得较高的识别率.

In the research on the facial images gender recognition method, there exist cases where a person attenos both training and testing, therefore, the conclusions attained are restrictive, In order to solve this problem, a mutu- ally independent testing set and training set have been established; the traditional gender recognition model is great- ly affected by the relevant parameters and the stability needs to be increased, therefore, a novel gender recognition method utilizing facial images and based on feature fusion has been proposed. With this method, the analysis of the main components and the orthogonal linear distinguishing analysis are combined to describe the overall features of the facial image, the restriction of order in binary classification for traditional linear distinguishing analysis is broken through, a balanced local binary value pattern method is applied to describe the local features of the image, and some of the overall features and the local features are fused to form the gender features of the facial image. The support vector machine is used for the classification of gender characteristics. The experimental results demonstrate that the proposed method ensures both a high recognition rate and robustness.