摘要
针对目前性别识别方法中的人体第二性特征提取困难、识别率低、鲁棒性差等问题,提出了一种基于神经网络的性别识别方法,并得到了基于人脸图像的性别识别分类器。文中先将人脸图像进行高斯滤波,再将预处理后的图像归一化用于训练BP神经网络,以得到性别识别分类器,最后将分类器与传统的性别识别方法进行比较。实验结果表明,通过文中方法实现了人体第二性特征自动提取,提高了分类器的容错能力和识别率,增强了鲁棒性。
This paper introduces the gender recognition method based on neural networks for the difficulty in ex-tracting the human secondary sex characteristics, and the low recognition rate and robustness. We get the gender recognition classifier based on human face images. The face image is processed by the Gauss filter, and then the normalization of preprocessed image is used to train the BP neural network to get the classifier. The gender recogni-tion method is compared with the traditional ones. The experimental results show that the extraction of the human secondary sex characteristics can be automatically done by our method, and the recognition rate, robustness and the fault tolerance are also improved.
出处
《电子科技》
2013年第9期151-154,共4页
Electronic Science and Technology
基金
山西省回国留学人员科研基金资助项目(2011-075)
山西省自然科学基金资助项目(2008011030)
关键词
性别识别
神经网络
高斯滤波
鲁棒性
gender recognition
neural networks
gaussian filter
robustness
