基于条件随机场的图像标注方法研究  

Research on Image Annotation Based on Conditional Random Field

作　　者：张扬睿 王卓燃 崔子琦 傅于窈 程世强 王伟[1] 白晨阳 ZHANG Yang-rui;WANG Zhuo-ran;CUI Zi-qi;FU Yu-yao;CHENG Shi-qiang;WANG Wei;BAI Chen-yang(School of Mechanical Electronic and Information Engineering,China University of Mining and Technology,Beijing 100083)

机构地区：[1]中国矿业大学(北京)机电与信息工程学院,北京100083

出　　处：《现代计算机》2021年第4期74-77,共4页Modern Computer

基　　金：中国矿业大学(北京)创新训练项目资助(No.C201904433)。

摘　　要：图像自动标注技术是实现图像检索的重要支撑。由于图像所包含信息的复杂性及“语义鸿沟”的存在,许多图像自动标注模型还无法达到高识别准确率的要求。如何挖掘标注标签之间的相关性是提高标注准确率的重要途径,介绍一种基于条件随机场的图像自动标注方法,该方法包含图像分割、特征提取、构建词袋模型、利用条件随机场模型实现图像自动标注等完整的步骤。The technology of Automatic Image Annotation(AIA)is an important support for image retrieval.Due to the complexity of image informa⁃tion and the existence of semantic gap,many AIA models cannot obtain high annotation accuracy.How to exploit the correlation among an⁃notation labels is an important way to improve the annotation accuracy.This work introduces an AIA method based on Conditional Random Field(CRF),it includes a complete process,including image segmentation,feature extraction,construction of word bag model,and the an⁃notation based on CRF.

关 键 词：图像自动标注 图像检索 条件随机场 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]