摘要
信息时代兴起的浪潮引发照片档案的数量成爆炸式的增长,本文针对照片档案进行检索利用存在的问题,分析了基于文本和基于内容的两种传统照片图像标注方式及优缺点,提出了一种多示例多标签照片档案自动标注技术,设计了基于图像归一化分割的仿射不变特征变换(SIFT)特征方法和多示例多标签学习训练成自动标注模型,阐述了多示例多标签照片档案自动标注主要流程步骤,实现档案照片自动标注及检索利用。
The rising tide of the information age has caused an explosive growth in the number of photo archives. This article focuses on the problems in the retrieval and utilization of photo archives, analyzes the advantages and disadvantages of two traditional photo image annotation methods based on text and content, and proposes An automatic labeling technology for multi-instance and multi-label photo archives. The affne invariant feature transform(SIFT)feature method based on normalized image segmentation is designed and multi-instance multilabel learning is trained into an automatic labeling model, and the multi-instance multi-label is explained. The main process steps of automatic labeling of photo files are realized to realize automatic labeling and retrieval of file photos.
作者
周文欢
ZHOU Wen-huan(Tianjin University Archives,Tianjin 300072)
出处
《山西档案》
2021年第4期124-129,共6页
Shanxi Archives
关键词
照片档案
自动标注
SIFT特征
多示例多标签学习
photo archives
automatic labeling
SIFT features
multi-instance multi-label learning
