基于人脸图像的性别识别技术  

The Techniques of Gender Recognition based on Face Images

作　　者：文环明[1] 王书海[1] 武瑞红[1] 

机构地区：[1]河北科技大学信息学院,石家庄050018

出　　处：《电脑开发与应用》2008年第1期25-27,共3页Computer Development & Applications

基　　金：河北省教育厅博士基金资助项目(B2004109)

摘　　要：以原始人脸图像为基础,用灰度拉伸、灰度均衡、中值滤波、图像锐化和图像变换等多种处理方法对人脸图像进行预处理,用分形几何的关联维方法提取人脸图像的特征值,并通过BP网络建立性别识别模型。在此基础上,用上述各种预处理方法对29个未参与建模的人脸图像处理,然后进行性别识别。结果表明,用灰度拉伸和傅立叶变换处理的人脸图像,性别识别效果较好;图像变换比其他处理方法的识别效果好;先对人脸图像作灰度拉伸,然后作傅立叶变换,性别识别效果最佳。On the basis of primitive face images, the face images are preprocessed with gray stretching, gray equalizing, median filtering, image sharpness and image transforms, while the character values of face images are abstracted through the correction dimension algorithm in fractals and the gender recognition models are made with the BP network. And then 29 face images without being modeled are processed with the above preprocessing methods before the gender recognition is taken. From the processed data, it shows that the gender recognitions through the gray stretching and Fourier transform of face images are better, and the gender recognitions of image transforms are better than that of other processing methods, and the gender recognition is best through the combination of gray stretching and Fourier transform of face images.

关 键 词：数字图像处理 性别识别 分形几何 关联维 人工神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]