摘要
为了克服欧式距离的度量方法在人脸特征表达上的不足,提出了一种基于KL距离的卷积神经网络人脸特征提取模型。通过卷积神经网络将输入样本转换为一个概率分布,利用KL距离度量不同样本之间概率分布的差异,并定义了一个代价函数对此距离进行优化,最后使用反向传播算法修改卷积神经网络的参数,使网络对人脸特征有更强的区分能力。将提取的特征向量通过神经网络分类器进行人脸验证,在YouTube等人脸库上进行了测试。试验结果表明,该方法不仅能提高正确率,而且还具有更好的泛化性能。
In order to overcome the shortcomings of Euclidean distance measurement in face feature expression, a neural network face feature extraction model based on KL divergence is proposed. The convolution neural network is used to transform the input sample into a probability distribution. The distance between different samples is measured by the KL divergence, and a cost function is defined to optimize the distance. The back propagation algo- rithm is used to modify the parameters of convolution neural network, the network has a stronger ability to distinguish between facial features. The extracted face feature vector is transformed into neural network classifier to performs face validation with YouTube face database. The experimental results show that the method can not only improve the error rate but also improve the generalization performance.
出处
《长沙理工大学学报（自然科学版）》
CAS
2017年第2期85-91,共7页
Journal of Changsha University of Science and Technology:Natural Science
基金
国家自然科学基金资助项目(11671125
71371065)
关键词
人脸识别
人脸验证
特征提取
KL距离
度量学习
卷积神经网络
face recognition
face verification
feature extraction
KL divergence
metric learning
convolutional neural network
