跨语图像检索中融合视觉信息的多语翻译与集成方法研究

项目名称： 跨语图像检索中融合视觉信息的多语翻译与集成方法研究

项目编号： No.61300077

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 黄永刚

作者单位： 北京理工大学

项目金额： 23万元

中文摘要： 跨语言图像检索可以使用户检索到与查询采用不同语言标注的图像，具有重要的研究和应用价值。针对跨语言图像检索中文本信息上下文缺乏的难题，本项目采用以图像视觉信息作为上下文以辅助多语翻译和集成的思想，研究一种融合图像视觉信息的多语翻译和集成新方法。首先，研究一种基于图像集合视觉相似性的查询翻译机制，提高查询翻译的准确率，从而改善目标语言图像检索的准确性。其次，在此基础上，研究一种基于检索性能预测和重排序的结果集成机制，实现对多个单语言检索结果的高效集成。最后，在结果集成之后，研究一种基于图像层次式聚类的图像标注翻译算法，提高对检索结果中图像标注的翻译准确率。项目将在中文标注图像集和英文标注图像集上验证方法的有效性。本项目的研究成果，对于跨语言图像检索、机器翻译等具有重要的理论和应用价值。

中文关键词： 查询翻译；图像检索；跨语检索；图像包；

英文摘要： Cross-Language Image Retrieval (CLIR) enables the users to retrieval the images annotated in different languages with the query. However, the lack of text information remains a challenge for multilingual translation and integration in CLIR. In order to address this issue, based on the idea of using the visual information as supplementary information, this project proposes a new method for multilingual translation and integration using visual information in CLIR. In the new method, firstly, a new query translation mechanism based on visual similarity of image sets is employed. The aim of the mechanism is to improve the accuracy of query translation and enhance the image retrieval performance in target languages. Secondly, in order to effectively integrate the image retrieval results in different languages, a new result integration mechanism based on prediction of the retrieval performance and re-ranking is adopted. Finally, a new algorithm for image annotation translation based on image hierarchical clustering is used, the goal of which is to enhance the performance of image annotation translation. The effectiveness of the proposed method will be demonstrated on the image sets annotated in Chinese and English. The achievements of this project have significant value in theory and practice for CLIR and Machine Tran

英文关键词： Query translation；Image retrieval；Cross language information retrieval；Image bag；