欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 视觉与图像 • 上一篇    下一篇

基于彩色人脸图像的亲子关系验证研究

  

  1. 1. 西北工业大学电子信息学院，陕西 西安 710072；
    2. 西安市蓝田县公安局刊事侦查大队，陕西 西安 710500；
    3. 西北大学信息科学与技术学院，陕西 西安 710127；
    4. 奥卢大学机器视觉与信号分析中心，芬兰 奥卢 9057
  • 出版日期:2017-10-31 发布日期:2017-11-03

Kinship Verification Through Color Face Images

  1. 1. School of Electronics and Information, Northwestern Polytechnical University, Xi’an Shaanxi 710072, China;
    2. The Criminal Investigation Brigade of the Public Security Bureau of LAN Tian County, Xi’an Shaanxi 710500, China;
    3. School of Information Science and Technology, Northwest University, Xi’an Shaanxi 710127, China;
    4. Center for Machine Vision and Signal Analysis, University of Oulu, Oulu Finland 90570
  • Online:2017-10-31 Published:2017-11-03

PDF (PC)

394

摘要： 人脸亲子关系验证是近几年新兴的并具有较大挑战性的研究方向，其通过分析两幅
人脸图像的面部特征判断是否具有亲子关系。针对现有亲子关系验证方法大都基于灰度人脸图像
展开，而忽略彩色图像色度信息的问题，通过提取颜色纹理特征，研究了色度信息在亲子关系验
证方面的影响，对比彩色空间与灰度空间实验结果，并分析了不同颜色空间对亲子关系验证的影
响。实验进一步研究了基于不同颜色空间融合的特征提取方法在亲子关系验证上的应用。
KinFaceW 和 TSKinFace 数据集上的实验结果表明，彩色图像可以取得较灰度图像更好的实验结
果，并且 HSV 颜色空间识别准确率最高。此外，融合 HSV 颜色空间和 YCbCr 颜色空间的特征
提取算法可以进一步提高识别效果。

关键词: 亲子关系验证, 人脸分析, 颜色纹理特征, 颜色空间

Abstract: Face-based kinship verification is a challenging and new problem in recent years, which aims
to automatically predict whether two persons have a biological kin relation or not by examining their
facial attributes. Based on the fact that most of the existing methods focus on gray-scale facial images,
thus discarding the chrominance information, this paper investigates the role of color in the verification
of kinship relationships from facial images. In this context, we analyze the effect of chrominance
information on kinship verification and compare the performance of kinship verification from color
images and gray images. Furthermore, our experiments involve in different color spaces and the
combination of weighted features from two color spaces. Experiment results on KinFaceW and
TSKinFace indicate that color images can achieve better performance than gray scale features and HSV
color space performs best. Moreover, this paper also discusses a feature extraction by fusing with
different color spaces. Fusing HSV with YCbCr color spaces can further improve the performance.

Key words: kinship verification, face analysis, color texture features, color spaces