欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
提出了一种基于两类核Fisher鉴别分析（KFDA）的人脸识别方法，对每2个不同人脸类别求解一个核Fisher鉴别函数，其优点是能针对特定的2个人脸图像类别，抽取区分该2类人脸的最佳鉴别特征，克服了多类KFDA和2类KFDA相比是次优的问题．为解决KFDA计算量大的问题，将MSE推广为基于核的MSE（KMSE），用其得到核Fisher鉴别函数，减少了训练和识别的计算时间．在识别阶段应用了两种融合方法融合各个基于KMSE的核Fisher鉴别函数．
英文摘要：
Multiclass Fisher discriminant analysis （MFDA） can extract Fisher optimal discriminant features. However MFDA is suboptimal compared with two-class FDA. To overcome the problem of MFDA, a face recognition method based on two-class FDA is proposed in this paper. This method can extract nonlinear Fisher optimal discriminant features distinguishing two specific classes of face images. Due to FLDA equivalent to minimum squared error （MSE） method, kernel based MSE is proposed to obtain kernel Fisher discriminant functions, in order to reduce the computational complexity in two-class FDA. In recognition stage, two fusion strategies to combine nonlinear Fisher discriminant functions are employed. Experiment results on ORL and Yale indicate that proposed two-class KFDA performs better than multiclass KFDA.
同期刊论文项目
同项目期刊论文
期刊信息
位置：