Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (23): 226-233.DOI: 10.3778/j.issn.1002-8331.2102-0330

• Graphics and Image Processing • Previous Articles     Next Articles

Kinship Verification Method of Facial Local Feature Enhancement

ZHENG Liang, CHEN Ya, CHEN Xiaopan, ZHENG Fengbin   

  1. School of Computer and Information Engineering, Henan University, Kaifeng, Henan 475004, China
  • Online:2021-12-01 Published:2021-12-02

人脸局部特征增强的亲属关系验证方法

郑亮,陈亚,陈小潘,郑逢斌   

  1. 河南大学 计算机与信息工程学院,河南 开封 475004

Abstract:

Kinship verification is an important branch of face recognition. The important applications of kinship verification include occasions such as search for missing children and relatives, family graph construction, social media analysis, and so on. Generally, there is a huge difference between facial images of parent and child. Therefore, how to extract discriminative features from facial images is the key to improve accuracy of kinship verification. In this paper, it proposes a method based on deep learning network to extract information of local facial regions by constructing a Local Facial Feature Enhancement Verification Net (LFFEV Net), which can obtain the face feature representation with strong discrimination for kinship verification. LFFEV Net consists of local feature attention network and residual verification network. The local feature attention network extracts the feature map of the key local facial area, and then inputs the obtained feature map and the original image together into the residual verification network to obtain the facial features with better discrimination. Finally, the features are fused and the Family ID information is combined to verify the kinship. Extensive experiments are conducted on the publicly available kinship dataset KinFaceW-I and KinFaceW-II. The experimental results demonstrate that the method can achieve high accuracy.

Key words: deep learning, kinship verification, local feature enhancement, attention network

摘要:

亲属关系验证是人脸识别的一个重要分支,可以用于寻找失散亲人、搜寻走失儿童、构建家庭图谱、社交媒体分析等重要场景。父母和孩子的人脸图像之间往往存在较大的差异,如何从人脸中提取到有鉴别力的特征是提高亲属关系验证准确率的关键。因此,提出了一种基于深度学习和人脸局部特征增强的亲属关系验证方法,构建了人脸局部特征增强验证网络(Local Facial Feature Enhancement Verification Net,LFFEV Net),获取用于亲属关系验证的具有强鉴别力的人脸特征表示。LFFEV Net由局部特征注意力网络和残差验证网络两部分组成。局部特征注意力网络提取人脸局部关键特征,将获取的局部关键特征和对应的原始图像一同输入到残差验证网络中获取更具鉴别力的人脸特征,将特征经过融合并结合Family ID信息进行亲属关系验证。算法在公开的亲属关系数据集KinFaceW-I和KinFaceW-II上进行测试,实验结果表明,所设计的方法在亲属关系验证任务中有较高的识别率。

关键词: 深度学习, 亲属关系验证, 局部特征增强, 注意力网络