正在加载图片...
第13卷第1期 智能系统学报 Vol.13 No.I 2018年2月 CAAI Transactions on Intelligent Systems Feb.2018 D0:10.11992/tis.201707018 网络出版地址:http:/kns.cnki.net/kcms/detail/23.1538.TP.20180116.1749.002.html 一种多层特征融合的人脸检测方法 王成济2,罗志明2,钟准2,李绍滋2 (1.厦门大学智能科学与技术系,福建厦门361005;2.厦门大学福建省类脑计算技术及应用重,点实验室,福建厦门 361005) 摘要:由于姿态、光照、尺度等原因,卷积神经网络需要学习出具有强判别力的特征才能应对复杂场景下的人脸检 测问题。受卷积神经网络中特定特征层感受野大小限制,单独一层的特征无法应对多姿态多尺度的人脸,为此提出 了串联不同大小感受野的多层特征融合方法用于检测多元化的人脸:同时,通过引入加权降低得分的方法,改进了目 前常用的非极大值抑制算法,用于处理由于遮挡造成的相邻人脸的漏检问题。在FDDB和WiderFace两个数据集上 的实验结果显示,文中提出的多层特征融合方法能显著提升检测结果,改进后的非极大值抑制算法能够提升相邻人 脸之间的检测准确率。 关键词:人脸检测;多姿态;多尺度;遮挡:复杂场景;卷积神经网络:特征融合;非极大值抑制 中图分类号:TP391.41文献标志码:A文章编号:1673-4785(2018)01-0138-09 中文引用格式:王成济,罗志明,钟准,等.一种多层特征融合的人脸检测方法小.智能系统学报,2018,13(1):138-146. 英文引用格式:WANG Chengji,.LUO Zhiming,ZHONG Zhun,ctal.Face detection method fusing multi-layer features.CAAl transactions on intelligent systems,2018,13(1):138-146. Face detection method fusing multi-layer features WANG Chengji,LUO Zhiming 2,ZHONG Zhun'2,LI Shaozi (1.Intelligent Science&Technology Department,Xiamen University,Xiamen 361005,China;2.Fujian Key Laboratory of Brain-in- spired Computing Technique and Applications,Xiamen University,Xiamen 361005,China) Abstract:To address the issues of pose,lighting variation,and scales,convolutional neural networks(CNNs)need to learn features with strong discrimination handle the face detection problem in complex scenes.Owing to the size limita- tions of the specific feature layer's receptive field in convolutional neural networks,the features computed from a single layer of the CNNs are incapable of dealing with faces in multi poses and multi scales.Therefore,a multi-layer feature fusion method that is realized by fusing the different sizes of receptive fields is proposed to detect diversified faces. Moreover,via introducing the method of weighted score decrease,the present usual non-maximum suppression al- gorithm was improved to deal with the detection omission of neighboring faces caused by shielding.The experiment res- ults with the FDDB and WiderFace datasets demonstrated that the fusion method proposed in this study can signific- antly boost detection performance,while the improved non-maximum suppression algorithm can increase the detection accuracy between neighboring faces. Keywords:face detection;multi pose;multi scale;occlude;complex scenes;convolutional neural network;feature fu- sion;non-maximum suppression 人脸识别技术作为智能视频分析的一个关键环着广泛的应用。人脸检测是人脸识别的基础关键环 节,在视频监控、网上追逃、银行身份验证等方面有 节之一,在智能相机、人机交互等领域也有着广泛 收稿日期:2017-07-10.网络出版日期:2018-01-18. 的应用。人脸检测是在输入图像中判断是否存在人 基金项目:国家自然科学基金项目(61572409,61402386,81230087, 脸,同时确定人脸的具体大小、位置和姿态的过 61571188). 通信作者:李绍滋.E-mail:szig@xmu.edu.cn. 程。作为早期计算机视觉的应用之一,人脸检测的DOI: 10.11992/tis.201707018 网络出版地址: http://kns.cnki.net/kcms/detail/23.1538.TP.20180116.1749.002.html 一种多层特征融合的人脸检测方法 王成济1,2,罗志明1,2,钟准1,2,李绍滋1,2 (1. 厦门大学 智能科学与技术系,福建 厦门 361005; 2. 厦门大学 福建省类脑计算技术及应用重点实验室,福建 厦门 361005) 摘 要:由于姿态、光照、尺度等原因,卷积神经网络需要学习出具有强判别力的特征才能应对复杂场景下的人脸检 测问题。受卷积神经网络中特定特征层感受野大小限制,单独一层的特征无法应对多姿态多尺度的人脸,为此提出 了串联不同大小感受野的多层特征融合方法用于检测多元化的人脸;同时,通过引入加权降低得分的方法,改进了目 前常用的非极大值抑制算法,用于处理由于遮挡造成的相邻人脸的漏检问题。在 FDDB 和 WiderFace 两个数据集上 的实验结果显示,文中提出的多层特征融合方法能显著提升检测结果,改进后的非极大值抑制算法能够提升相邻人 脸之间的检测准确率。 关键词:人脸检测;多姿态;多尺度;遮挡;复杂场景;卷积神经网络;特征融合;非极大值抑制 中图分类号:TP391.41 文献标志码:A 文章编号:1673−4785(2018)01−0138−09 中文引用格式:王成济, 罗志明, 钟准, 等. 一种多层特征融合的人脸检测方法[J]. 智能系统学报, 2018, 13(1): 138–146. 英文引用格式:WANG Chengji, LUO Zhiming, ZHONG Zhun, et al. Face detection method fusing multi-layer features[J]. CAAI transactions on intelligent systems, 2018, 13(1): 138–146. Face detection method fusing multi-layer features WANG Chengji1,2 ,LUO Zhiming1,2 ,ZHONG Zhun1,2 ,LI Shaozi1,2 (1. Intelligent Science & Technology Department, Xiamen University, Xiamen 361005, China; 2. Fujian Key Laboratory of Brain-in￾spired Computing Technique and Applications, Xiamen University, Xiamen 361005, China) Abstract: To address the issues of pose, lighting variation, and scales, convolutional neural networks (CNNs) need to learn features with strong discrimination handle the face detection problem in complex scenes. Owing to the size limita￾tions of the specific feature layer’s receptive field in convolutional neural networks, the features computed from a single layer of the CNNs are incapable of dealing with faces in multi poses and multi scales. Therefore, a multi-layer feature fusion method that is realized by fusing the different sizes of receptive fields is proposed to detect diversified faces. Moreover, via introducing the method of weighted score decrease, the present usual non-maximum suppression al￾gorithm was improved to deal with the detection omission of neighboring faces caused by shielding. The experiment res￾ults with the FDDB and WiderFace datasets demonstrated that the fusion method proposed in this study can signific￾antly boost detection performance, while the improved non-maximum suppression algorithm can increase the detection accuracy between neighboring faces. Keywords: face detection; multi pose; multi scale; occlude; complex scenes; convolutional neural network; feature fu￾sion; non-maximum suppression 人脸识别技术作为智能视频分析的一个关键环 节,在视频监控、网上追逃、银行身份验证等方面有 着广泛的应用。人脸检测是人脸识别的基础关键环 节之一,在智能相机、人机交互等领域也有着广泛 的应用。人脸检测是在输入图像中判断是否存在人 脸,同时确定人脸的具体大小、位置和姿态的过 程。作为早期计算机视觉的应用之一,人脸检测的 收稿日期:2017−07−10. 网络出版日期:2018−01−18. 基金项目:国家自然科学基金项目 (61572409, 61402386, 81230087, 61571188). 通信作者:李绍滋. E-mail:szlig@xmu.edu.cn. 第 13 卷第 1 期 智 能 系 统 学 报 Vol.13 No.1 2018 年 2 月 CAAI Transactions on Intelligent Systems Feb. 2018
向下翻页>>
©2008-现在 cucdc.com 高等教育资讯网 版权所有