基于多层字典学习的传统服饰图像标注算法  

Traditional clothing image annotation algorithm based on multi-layer dictionary learning

作　　者：王梓舟 赵海英 任文超 WANG Zizhou;ZHAO Haiying;REN Wenchao(School of Computer,Beijing University of Post and Telecommunication,Beijing 100876,China;Artificial Intelligence Institute,Beijing University of Post and Telecommunication,Beijing 100876,China)

机构地区：[1]北京邮电大学计算机学院,北京100876 [2]北京邮电大学人工智能学院,北京100876

出　　处：《中国传媒大学学报（自然科学版）》2022年第4期2-7,18,共7页Journal of Communication University of China：Science and Technology

基　　金：揭榜挂帅重点研发课题(课题编号:2021YFF0901701)。

摘　　要：中国传统服饰图像是中华优秀传统文化重要的组成部分,图像内涵丰富,时空跨度大,但理解存在歧义,急需一套语义解读方法,而大量图像标注算法主要关注在各自的垂直领域,传统服饰图像仍然面临着标注精度亟需提高的挑战。本文以中国传统服饰图像作为研究对象,以字典学习多标签标注方法作为研究方法,提出了融合深度多层结构框架的多标签字典学习算法,通过结合字典学习与多层结构框架来提高标注性能。最后通过对比实验验证了该思路的正确。Chinese traditional clothing image is an important part of Chinese excellent traditional culture.The image connotation is rich,the space-time span is large,and the understanding is ambiguous.There is an urgent need for a set of semantic interpretation methods for it.A large number of image annotation algorithms mainly focus on their respective vertical fields,and traditional clothing images are still facing the challenge of improving the annotation accuracy.This paper takes Chinese traditional clothing images as the research object,takes dictionary learning multi label tagging method as the research method,and proposes a multi label dictionary learning algorithm integrating deep multi-layer structure framework,which improves the tagging performance by combining dictionary learning and multi-layer structure framework.Finally,the correctness of this idea is verified by comparative experiments.

关 键 词：传统服饰图案 多标签标注 多层字典学习 字典相关性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]