Current Issue Cover


摘 要
摘 要:目的 人脸图像分析是计算机视觉和模式识别领域的重要研究方向之一,基于人脸图像的血缘关系识别是对给定的一对或一组人脸图像,判断其是否存在某种血缘关系。人脸血缘关系识别不仅在生物特征识别领域有着重要研究价值,而且在社交媒体挖掘、失散家庭成员寻找等社会生活领域中有重要的应用价值。针对当前大多数算法都是基于传统机器学习方法,提出一种采用深度度量学习进行人脸图像血缘关系研究的新方法。方法 目前深度学习算法能很好的理解单张人脸图像,但是多个主体间的关系探究仍然是计算机视觉领域富有挑战性的问题之一。为此,提出一种基于深度度量学习的父母与子女的血缘关系识别方法。首先使用超过5,000,000张人脸图像的样本集训练一个深度卷积神经网络FaceCNN并提取父母与子女的人脸图像深度特征,之后引入判别性度量学习方法,使得具有血缘关系的特征尽可能的靠近,反之则尽可能远离。然后对特征进行分层非线性变换使其具有更强判别特性。最后根据余弦相似度分别计算父亲、母亲和孩子的相似度并利用相似概率值得到双亲和孩子的综合相似度得分。结果 算法在TSKinFace数据集上验证了FaceCNN提取特征与深度度量学习结合进行血缘关系识别的有效性,最终在该数据集上父母与儿子和女儿的血缘关系识别准确率分别达到87.71%和89.18%,同时算法在进行血缘度量学习和双亲相似度计算仅需要3.616秒。结论 提出的血缘关系识别方法,充分利用深度学习网络良好的表征和学习能力,不仅耗时少,而且有效地提高了识别准确率。
Kinship verification combined with deep metric learning

Qu Daoqing,Ma Lin,Shao Zhuhong(Capital Normal University)

Abstract: Objective Facial image analysis has been an important and active subfield in computer vision, and pattern recognition. Kinship verification refers to the tasks of training a machine, based on features extracted from facial images, to determine the existence, type and degree of kinship for two or more people. Kinship verification has wide and promising applications in many technical and social fields, such as family history researches and photo management in social network services. In addition, the study of Kinship verification is of great theoretical significance to related fields,Several commonalities can be observed among kinship verification algorithms,which are based on traditional machine learning,this paper proposes a new method to study kinship verification based on face images by deep metric learning. Method Currently the deep learning algorithm can well understand a single face image, but the relationship between multiple subjects is still one of challenging problems in the field of computer vision. This paper presents a kinship verification approach between parents and their children. First of all, a deep convolution neural network “FaceCNN” is trained by using a data set containing more than 5,000,000 face images, where the deep features of face images are extracted. The training set of the deep convolutional neural network contains of thousands of people who mostly comes from the internet,including the public kinship data set,microsoft face data sets,some films and television dramas,public figures,each of which contains of dozens of face image. These face images have complex backgrounds,large differences in image quality,various face postures, expressions, ages, and genders. Before training the deep network, face images in the data set need to be preprocessed, including face filtering, face detection, and key point alignment. Then, all images are cut into “144×144” pixels. Afterwards, we divide the samples father-mother-son(FMS)and father-mother-daughter(FMD)into father-son(FS),mother-son(MS),father-daughter(FD)and mother-daughter(MD). But we cannot use the Mahalanobis distance to perform metric learning for kinship verification because the traditional Mahalanobis distance measurement learning method only seeks a linear transformation and does not capture the non-linear manifold of the face image well. A discriminative metric learning algorithm called DDML is followed to make the kinship features be as close as possible and vice versa. The hierarchical non-linear transformation is employed to achieve more discriminative. Finally, a method of kinship verification for parents and children is proposed,which the cosine similarity is calculated between parents and children. Children may be more similar to one of the parents in daily life,so that we must calculate the similar probability of the children with their father and mother respectively. The final kinship score is a weighted average of the similarity between children and parents. Result A nine-layer deep model was trained to determine whether the given images have a relationship. The effectiveness of extraction features with “FaceCNN” combined with deep metric learning for kinship verification is validated in the experiment,which the recognition accuracy between parents and son performed on TSKinFace dataset is 87.71%, and that between parents and daughter is 89.18%, which recognition rate of the model increased by 1.31% and 4.87%. Meanwhile, the proposed approach only needs 3.616 seconds for metric learning and parental similarity calculation reducing time consuming greatly compared with other algorithms. Conclusion The proposed kinship verification takes full advantage of the characterization and learning ability of deep neural network, not only consumes less time but also improves the recognition accuracy efficiently.