基于深度学习的实时视频通话人像抠图方法  

Real-time video ting method based on deep learning

作　　者：李曈 LI Tong(School of Computer Science and Engineering,Northeastern University,Shenyang 110819,China)

机构地区：[1]东北大学计算机科学与工程学院,辽宁沈阳110819

出　　处：《无线互联科技》2024年第11期104-106,共3页Wireless Internet Technology

摘　　要：文章提出基于深度学习的实时视频通话人像抠图方法,充分利用了多维监控空间进行抠图特征多维识别,从而更加准确地提取人像边缘和背景特征,进一步优化抠图效果。研究通过生成人像三分图,建立适用于实时视频通话的人像抠图模型,为人像编辑提供了更高质量的抠图工具。实验结果显示,这一方法不仅在抠图效果上表现出较高的连接度,而且具有较好的应用潜力,可以有效提升人像抠图的精准性和实时性,对视频通话等场景具有积极意义。The article proposes a real-time video call portrait cutout method based on deep learning,which fully utilizes the multidimensional monitoring space for multidimensional recognition of cutout features,thereby achieving more accurate extraction of portrait edges and background features,and further optimizing the cutout effect.A portrait cutout model suitable for real-time video calls was established by generating a portrait tripartite graph,providing a higher quality cutout tool for portrait editing.The experimental results show that this method not only exhibits high connectivity in the cutout effect,but also has good application potential,which can effectively improve the accuracy and real-time performance of portrait cutout,and has positive significance for scenarios such as video calls.

关 键 词：深度学习 视频通话 人像抠图 抠图方法 

分 类 号：G642[文化科学—高等教育学]