性别鉴别的线性可分性分析

Linear separability of gender classification

首次从线性可分性的角度探讨了人脸图像的性别鉴别问题。通过对常用线性与非线性特征抽取方法以及一类改进的非线性特征抽取方法的对比分析及不同情况下性别鉴别的实验对比,较全面地考察了各种特征抽取方法所对应的数据的线性可分性及分类效果。首次提出从人脸肤色等角度考虑人脸图像的性别鉴别问题,并给出了指示意义较强的鉴别方法与方案建议。

In this paper,the authors first analyze linear separability of face images for gender classification.The authors compare the popular linear feature extraction,nonlinear feature extraction,and a reformative nonlinear feature extraction method,and conduct gender classification experiments using these methods under different conditions.These experiments clearly reveal linear separability and classification performance of the feature extraction results obtained using different methods.The authors are also the first to state that gender classification issue should take complexion into account.Additionally,the authors propose a novel significant gender classification strategy and scheme.