摘要
性别分类是指根据人脸部分的图像判别其性别的模式识别问题。探讨使用头发信息作为特征进行基于人脸图像的性别分类,提出了一种检测正面人脸图像中头发区域的方法,定义了6种头发特征并且提出了相应的特征提取方法。通过在两个人脸库上的对比实验,发现相对于特征脸(PCA)、Fisher脸(LDA)仅仅作用于人脸内部的特征提取方法,使用头发作为特征能使性别分类的平均准确率提高2.7%～8.2%。该实验结果说明了头发特征对于性别分类的重要性。
Gender Classification refers to judge people＇s gender based on their facial images. This paper discusses how to use hair features to solve gender classification problems and proposes a method for detecting hair area on the front view facial image. It defines six different hair features to represent hair and introduces the corresponding approaches to calculate them. The proposed hair features are compared with the traditional Eigen - Face （PCA） and Fisher - Face （LDA） features on two face image databases. Experimental result indicates that the proposed hair features are superior to both Eigen - Face （PCA） and Fisher - Face （LDA） features and obtain 2. 7% to 8.2% accuracy improvements on the average. This demonstrates that hair features are of great importance to gender classification.
出处
《计算机仿真》
CSCD
北大核心
2009年第2期212-216,共5页
Computer Simulation
基金
国家自然科学基金(60473040)
