激光与光电子学进展, 2019, 56 (20): 201002, 网络出版: 2019-10-22  

基于人脸聚类的视频中人脸图像优选方法的研究 下载: 1249次

Research on Face Image Optimization Method Based on Face Clustering in Video
作者单位
延边大学工学院智能信息处理研究室, 吉林 延吉 133002
摘要
人脸图像优选对智能监控系统中的人脸识别有着重要的意义。针对在视频中多人脸跟踪时出现跟丢、跟错以及无法及时添加、取消跟踪器等难以处理的问题,本文提出用人脸聚类代替人脸跟踪获取同一人脸图像,并构造出一种人脸图像质量的综合评价指标来从大量的多姿态人脸图像中选出一张人脸姿态和图像质量较好的人脸图像。首先对视频中的行人进行人脸检测,然后采用残差网络提取人脸面部特征进行人脸聚类,最后定义了人脸旋转程度、人眼状态、人脸遮挡程度、人脸图像清晰度4个评价指标,并将聚类后每一类人脸图像在4个评价指标上归一化均值作为各评价指标的权重系数,从而构造出一种人脸图像质量的综合评价指标,以此进行人脸图像优选。实验结果表明,人脸聚类能够有效获取到视频中同一人的人脸图像,所构造的人脸图像质量综合评价指标能有效获取到视频中的同一人的较优人脸图像。
Abstract
Face image optimization has important significance for face recognition in intelligent monitoring system. In the case of multi-face tracking in video, there are problems such as tracking error and inability to add and cancel the tracker in time. This paper proposes a face clustering method replacing the face tracking method to obtain face images of the same person, and a face image quality evaluation method to select a face image with good face pose and good image quality from a large number of multi-pose face images of the same person. First, the face detection from the video frame is performed, and then the residual network is used to extract the facial features for face clustering. Finally, the normalized mean value is computed as the weight coefficient of corresponding evaluation index for each type of face after clustering. Consequently, a comprehensive evaluation index is constructed to optimize the face image. Experiments show that face clustering can effectively obtain the same face image, and the constructed face image quality comprehensive evaluation index can effectively select a better face image from the same face images.

王畅, 崔荣一, 金璟璇, 金小峰. 基于人脸聚类的视频中人脸图像优选方法的研究[J]. 激光与光电子学进展, 2019, 56(20): 201002. Chang Wang, Rongyi Cui, Jingxuan Jin, Xiaofeng Jin. Research on Face Image Optimization Method Based on Face Clustering in Video[J]. Laser & Optoelectronics Progress, 2019, 56(20): 201002.

引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!