基于深度卷积神经网络End-to-End模型的亲属关系认证算法

摘要

针对如何利用人脸图像进行亲属关系认证的问题,提出基于深度卷积神经网络End-to-End模型的亲属关系认证算法.首先,构建一个包含卷积层、全连接层和soft-max分类层的深度卷积神经网络模型.卷积层可以提取亲子图像的隐性特征,全连接层可以将提取的隐性特征映射为亲属关系认证的二分类问题,soft-max分类层可以直接判断该对样本是否具有亲属关系.然后,将成对的标记训练数据输入网络进行迭代,优化深度网络模型参数,直至损失曲线稳定.最后,利用训练完毕的深度网络模型对输入测试图像对进行分类判决,通过统计得到最终的准确率.在KinFaceWI和KinFaceWII数据库上的结果显示,相比以往的亲属关系认证算法,文中模型具有更好的性能.%An algorithm based on deep convolutional neural network end-to-end model is proposed to solve the problem of kinship verification with facial image. Firstly, a deep convolutional neural network model is constructed. It consists of convolutional layers, fully connected layer and soft-max classification layer. The implicit features of parent-child images can be extracted by convolution layers. Then, the extracted latent features can be mapped into two-class classification problem of kin verification by fully connected layer, and the kinship relationship of samples can be directly determined by the soft-max classifier. Then, the paired tag training data are inputted into the network to be iterated and parameters of the deep network model are optimized until the loss curve is stable. Finally, the input image pairs are classified by the trained parameters, and the final accuracy is obtained by statistics. The experimental results on the KinFaceW-I database and KinFaceW-II dataset demonstrate the proposed convolutional neural network end-to-end model outperforms other kinship verification algorithms.