基于主动学习的相关反馈算法研究

Research on Reference Feedback based on Active Learning

相关反馈(reference feedback)是信息检索领域中一种常用技术,近年来,该技术被广泛应用与基于内容的图像检索(CBIR)领域中,旨在通过用户与图像检索系统间的交互过程,克服图像底层特征与高层语义之间的语义鸿沟问题。将主动学习算法结合到相关反馈技术当中,其目的是利用主动学习算法,从无标记图像集中选择最具有信息化的部分图像作为反馈图像,减少用户与系统之间的反馈次数。在COREL图像库和VOC图像库上,对基于主动学习的相关反馈技术进行实验验证,实验结果证明了,基于主动学习的相关反馈技术可以有效提高图像检索系统的性能。

Reference feedback is a common technique in the field of information retrieval. Recently, it has been widely applied in the task of content - based image retrieval （CBIR） in order to overcome the gap between low features and high semantic concepts by utilizing the user - computer interaction. This paper combines the active learning into reference feedback technique. The goal is to choose the most informative images in the unlabeled pool as returned images by using the active learning, and then reduces the feedback times. On the COREL database and VOC database, the paper evaluates the proposed method. The experimental results indentify that the reference feedback based on active learning can effectively enhance the performance of CBIR system.