旋转人脸检测算法  

ROTATING FACE DETECTION ALGORITHM

作　　者：张立亮 滕国伟[1] 范涛[1] 李聪[2] Zhang Liliang;Teng Guowei;Fan Tao;Li Cong(Shanghai Institute for Advanced Communication and Data Science,School of Communication and Information Engineering,Shanghai University,Shanghai 200444,China;AVCON Information Technology Co.,Ltd,Shanghai 200438,China)

机构地区：[1]上海先进通信与数据科学研究院上海大学通信与信息工程学院,上海200444 [2]华平信息技术股份有限公司,上海200438

出　　处：《计算机应用与软件》2019年第12期137-142,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61571285)

摘　　要：人脸检测技术作为机器视觉领域的挑战性课题,近些年取得了很大发展,其中无约束环境下的平面内旋转人脸检测是难点之一,为业界所重点关注。目前的人脸检测大多基于区域候选和预选矩形框的设计,进行分类和回归,但这种方式只能回归得到目标人脸的旋转体,并不能反映目标的正确方位。提出一种改进的旋转矩形框的方法,设计符合人脸的旋转预选框,并借鉴水平预选框的人脸检测方法来构建网络。选择FDDB和Wider Face中人脸数据集进行统一的标签信息处理,使用该数据集训练和测试网络。实验结果表明,该方法具有旋转不变性,对旋转人脸检测模型具有较优的鲁棒性。As a challenging subject in the field of machine vision,face detection technology has made great progress in recent years.In-plane rotating face detection under unconstrained environment is one of the difficulties,and it is the focus of the industry.At present,face detection is mostly based on the design of regional proposal and prior-box for classification and regression.However,this method can only return to the rotating body of the target face,and cannot reflect the correct orientation of the target.This paper proposes an improved method of rotating a rectangular frame,which designs a rotating preselection box that conformed to the face,and draws on the face detection algorithm of the horizontal prior-box to construct the network.We selected the face dataset in FDDB and Wider Face for unified tag information processing,and used this dataset to train and test the network.Experimental results show that this model has rotation invariance and is robust to rotating face detection models.

关 键 词：预选框 人脸检测 旋转不变性 鲁棒性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]