基于样本互作转换的人脸识别

Face Recognition Based on Samples Interaction Transformation

针对人脸识别问题,因人脸图像具多姿态性、实际样本少等特性,且传统识别方法存在信息利用率低、分类过程繁琐等缺陷。为提高人脸识别精度并简化模型,提出了一种新的多人脸样本互作转换方法,将同时识别多个人脸的多分类问题简化为简单二分类问题,进行统一建模,充分利用图像信息,并基于可交换核函数消除互作样本中因初始样本排列顺序不同带来的影响,然后对产生的新特征进行非线性筛选,最后以简单投票策略对独立测试样本进行类别校正。基于ORL人脸数据库进行仿真,独立测试样本识别准确率高于95%,明显高于参比模型。实验结果表明,样本互作转换能有效简化识别模型,在海量数据识别中具有较好的应用价值。

In face recognition research, due to some problems of the face images such as multi-pose and small actual samples, the traditional identification methods have the defects of low information utilization rate and cumbersome classification process. To improve face recognition accuracy and simplify the model, a sample interaction transformation was proposed in this paper, which can transform the multi-classification to a simple binary class4fication, and has the trait of full use of image information. A symmetrical kernel function was inducted to solve the rank problem of the two initial samples in interaction sampling pair and irrelevant, and redundant features were eliminated nonlinearly with support vector machine. Finally, the prediction results were further corrected by simple-vote decision. The sim- ulation experiments based on ORL face database and the new method have the highest recognition accuracy in all ref- erence models. The result shows that interaction transformation can simplify the model and has a good value in recog- nition of the huge amounts of data.