WANG Chang,JIN Jingxuan*,JIN Xiaofeng.Research on face data set generation method based oncombination of clustering and tracking[J].Journal of Yanbian University,2019,45(03):221-227.
聚类与跟踪相结合的人脸数据集生成方法研究
- Title:
- Research on face data set generation method based on combination of clustering and tracking
- 文章编号:
- 1004-4353(2019)03-0221-07
- 分类号:
- TP391.4
- 文献标志码:
- A
- 摘要:
- 为提高人脸数据集生成的准确率和效率,提出了一种将人脸跟踪与人脸聚类相结合的人脸数据集生成方法.首先,对KCF算法进行改进,并将改进的KCF算法用于人脸跟踪,得到不同时间片段的人脸图像集; 其次,从每个人脸图像集中通过人脸图像优选算法选出高质量的人脸图像; 再次,将优选出来的人脸图像进行人脸聚类,以完成视频中每个人的人脸数据集的生成; 最后,通过实验对比基于人脸跟踪、基于人脸聚类和基于本文方法的人脸数据集生成效果.实验结果表明,本文方法与基于人脸跟踪的人脸数据集生成方法相比,纯度提升约15%; 与基于人脸聚类的人脸数据集生成方法相比,效率提升约50%.
- Abstract:
- To improve the accuracy and efficiency of face data set generation, we propose a method of combining face tracking and face clustering to generate face data sets. Firstly, the KCF algorithm is improved and used for face tracking to get face image sets with different time segments. Secondly, a high-quality face image is selected from each face image set by face image optimization algorithm. Thirdly, face clustering is carried out on all optimized face images to generate face data set for each person in the video. Finally, three different face data sets generation experiments of the method based on face tracking, the method based on face clustering and the method proposed in this paper are performed. The experimental results show, comparing with the method of face data sets generation based on face tracking algorithm, the method proposed in this paper has improved about 15% in purity, comparing with the method of face data sets generation based on face clustering algorithm, the method proposed in this paper has improved about 50% in time efficiency.
参考文献/References:
[1] 李蕊岗,张明.基于改进Camshift和Kalman滤波的目标跟踪算法[J].微型机与应用,2017,36(12):39-41.
[2] HENRIQUES J F, CASEIRO R, MARTINS P, et al. High-speed tracking with Kernelized Correlation Filters[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015,37(3):583-596.
[3] NAM Hyeonseob, HAN Bohyung. Learning multi-domain convolutional neural networks for visual tracking[C]//Computer Vision and Pattern Recognition. Las Veges, USA: CVPR, 2016:4293-4302.
[4] SAHOO S K, MAKUR A. Dictionary training for sparse representation as generalization of K-means clustering[J]. IEEE Signal Processing Letters, 2013,20(6):587-590.
[5] 陈恳.人脸深度特征获取及聚类技术研究[D].杭州:浙江工业大学,2017.
[6] 王玉晗,罗邓三郎.聚类算法综述[J].科技资讯,2018,16(24):10-11.
[7] BIEMANN Chris. Chinese whispers: an efficient graph clustering algorithm and its application to natural language processing problems[C]//The First Workshop on Graph Based Methods for Natural Language Processing. Stroudsburg: Association for Computational Linguistics, 2006:73-80.
[8] 张乘龙,夏筱筠,柏松,等.基于KCF跟踪算法的目标轨迹记录系统[J].计算机系统应用,2017,26(5):113-118.
[9] HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR). New York: IEEE, 2016:770-778.
备注/Memo
收稿日期: 2019-04-23 *通信作者: 金璟璇(1972—),女,副教授,研究方向为计算机视觉、智能算法等.
基金项目: 吉林省教育厅“十三五”科学技术项目(JJKH20191126KJ); 延边大学世界一流学科建设培育项目(18YLPY14)