基于内容的图像检索系统性能评价

在图像检索需求多元化和专业化的推动下，CBIR技术日趋成熟，目前已有越来越多的商用和科研系统相继推出。这样就迫切需要展开对CBIR系统性能评价标准的研究，因为任何一项技术都是由该领域中相应的评价标准来推动的。为了使人们对这方面的现状动态有一概略了解，首先讨论了基于内容的图像检索评价过程中的两个基本问题，即大规模数据库的建立和获取进行相关性评判，然后对近年来文献中所见的基于内容的图像检索系统性能评价方法进行了回顾和综述；最后在此基础上，对其发展方向进行了探讨，并提出建立一个标准的测试数据集用来推动基于内容的图像检索系统性能评价的发展和更好地将人结合到基于内容的图像检索系统的性能评价过程中，以发展交互式的性能评价方法的建议。

An Overview of Performance Evaluation in Content-based Image Retrieval

Promoted by the professional and diverse demands of image retrieval, the technique of content-based image retrieval (CBIR) becomes more mature. More and more commercial and scientific research systems are developed. As any technique is promoted by the performance evaluation of corresponding research area, for the development of effective image retrieval applications it is imperative to study the standard of performance evaluation in content-based image retrieval. Problems such as a common image database for performance comparison and a means of getting correlation judgement for queries are explained. This paper presents a review of the methods of performance evaluation in content-based image retrieval proposed in the literatures and tries to figure out the developing direction in the future. This paper also recommends that the content-based retrieval research community should establish a standard test-bed for evaluating image retrieval effectiveness. Further work needs to be done to better involve users in the evaluation process because the ultimate aim is to measure the usefulness of a system for a user. Interactive performance evaluations including several levels of feedback and user interaction need to be developed.