基于人脸尺度约束的人像自动抠图方法  

AUTOMATIC HUMAN MATTING METHOD BASED ON FACE SCALE CONSTRAINT

作　　者：黄叶珏[1] 潘翔[2] 褚一平 Huang Yejue;Pan Xiang;Chu Yiping(Department of Information Technology,Zhejiang Economic&Trade Polytechnic,Hangzhou 310018,Zhejiang,China;College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310023,Zhejiang,China;Hangzhou Mass Information Technology Co.,Ltd.,Hangzhou 310012,Zhejiang,China)

机构地区：[1]浙江经贸职业技术学院信息技术系,浙江杭州310018 [2]浙江工业大学计算机科学与技术学院,浙江杭州310023 [3]杭州海量信息技术有限公司,浙江杭州310012

出　　处：《计算机应用与软件》2022年第1期158-162,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(61871350)。

摘　　要：人像抠图算法是许多人像图像处理方法的核心,而人像三分图的准确性直接影响抠图的效果。提出一种通过关键点估计人脸尺度,根据尺度控制三分图生成网络产生标准化的人像三分图,从而提高人像抠图结果的方法。同时构建一个包含19118幅人像的数据集用于训练和测试模型,并提出一种多级微调的方式来训练模型,以降低训练难度并获得更好的效果。实验结果表明,在人像占图像比例不可控的场景中,该方法可以取得较好的结果。The human matting algorithm is the core of human image processing,and the accuracy of the portrait trisection directly affects the matting results.This paper proposes a method to estimate the face scale through key landmarks,and generate a standardized human trisection map by controlling the trisection map generation network according to the scale,which improves the results of human matting.We constructed a dataset containing 19118 human matting images for training and testing,and put forward a multi-level finetuning method to train the model,so as to reduce the training difficulty and obtain better results.The experimental results show that the method can achieve better results in the scene where the proportion of the portrait in the image is uncontrolled.

关 键 词：人像抠图 人脸关键点 语义分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]