Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (21): 36-45.DOI: 10.3778/j.issn.1002-8331.1908-0308

Previous Articles     Next Articles

Survey on Social Relationship Recognition Based on Images

GAO Jianjun, QING Linbo   

  1. School of Electronics and Information Engineering, Sichuan University, Chengdu 610065, China
  • Online:2019-11-01 Published:2019-10-30



  1. 四川大学 电子信息学院,成都 610065

Abstract: Social relationship is the general term for the relationship between people in the process of common material and spiritual activities. So far, all the researches have been done before are all related to the fields like social science, face recognition, impression recognition, motion recognition, scene recognition, object detection. As for this paper, start with a definition of different categories of relationships, it is divided into 4 parts from simple relation to difficult one by summarizing the researches relating to the topic. It introduces the method to extract the features and classification based on both traditional machine learning and deep learning. Meanwhile, the paper concludes the datasets and mechanism for different categories and compares the results, advantages and disadvantages of every work. This paper discusses the utilization of relationship recognition in real world and the possible direction that can be studied in the future.

Key words: social relationship, relationship recognition, computer vision, machine learning, image classification

摘要: 社会关系是人们在共同的物质和精神活动过程中所结成的相互关系的总称。目前已有相关的工作对其进行了研究,其中涉及到关系社会学、人脸识别、表情识别、动作识别、场景识别和物体检测等相关领域。从不同划分方法下的社会关系出发,通过总结相关研究将社会关系从简单到复杂,将其划分成4类:kinship、groups、activities & interactions和detailed relationship。阐述了对于几种不同关系划分在识别时所用到的特征提取及其分类方法,主要分为传统机器学习和深度学习两个模块。然后对不同的模型所使用到的相关数据集和机制进行了介绍,并对各个模型的结果、优缺点和适用范围进行了分析,最后对社会关系识别未来的研究方向及应用前景进行了探讨。

关键词: 社会关系, 关系识别, 计算机视觉, 机器学习, 图像分类