Exploiting Multimodel Synergy for Large Scale and Diverse Image Retrieval in Digital Archives

利用多模型协同作用进行数字档案中的大规模和多样化图像检索

• 批准号: 0535162
• 负责人: Zhongfei Zhang
• 金额: --
• 依托单位: SUNY at Binghamton
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2006
• 资助国家: 美国
• 起止时间: 2006-04-15 至 2010-03-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0535162&HistoricalAwards=false
• 关键词: Exploiting; Multimodel; Synergy; Large; Scale

The goal of this project is advance the state-of-the-art in image retrieval by explicitly addressing two issues: the semantic gap and the scalability. The semantic gap is related to the retrieval effectiveness, e.g., being able to use natural language to describe the images of interest, rather than having to specify low-level features, such as color histograms. The scalability refers to both the diversity in image content and quality and the number of images in the database. The specific approach consists of using probabilistic models and novel co-clustering techniques to explicitly exploit the synergy between multiple modalities typically co-existing in many real world image retrieval applications. Broader impact includes a series of innovative community outreach activities. The results of this research project will be incorporated into the development of image search engines and K-12 learning tools in two non-profit education and service organizations. The specific scenarios include a museum and a botanical garden, with a direct benefit to the communities and the society these organizations provide services to. Knowledge dissemination includes providing learning and research experience to students involved in the project. In addition, the organizations' technical personnel will provide expertise to develop useful services to the communities and society, will help in validating the results and also benefit from involvement in the project. The project Web site (http://www.fortune.binghamton.edu/ir.htm) will be used to disseminate further information and results.

该项目的目标是通过明确解决两个问题：语义差距和可扩展性来推进图像检索的最新发展。语义鸿沟与检索的有效性有关，例如，能够使用自然语言来描述感兴趣的图像，而不必指定低级特征，例如颜色直方图。可伸缩性是指图像内容和质量的多样性以及数据库中图像的数量。具体的方法包括使用概率模型和新的共聚类技术，明确利用通常共存于许多真实的世界的图像检索应用程序之间的协同作用。更广泛的影响包括一系列创新的社区外联活动。该研究项目的成果将被纳入两个非营利性教育和服务机构的图像搜索引擎和K-12学习工具的开发中。具体方案包括博物馆和植物园，这些组织提供服务的社区和社会直接受益。知识传播包括向参与项目的学生提供学习和研究经验。此外，各组织的技术人员将提供专门知识，为社区和社会提供有益的服务，帮助验证结果，并从参与项目中受益。该项目的网址（http：//www.fortune.binghamton.edu/ir.htm）将用于传播进一步的信息和结果。