【作者】 宫明营； 王智愚； 孙立峰； 杨士强；

【Author】 Mingying GONG , Zhiyu WANG, Lifeng SUN, Shiqiang YANG Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China

【机构】 清华大学计算机科学与技术系；

【摘要】 随着互联网图片量的急剧增加,人们获取图片越来越依赖于计算机的自动处理。相似图像搜索作为通用图像搜索的一个扩展,是当前多媒体领域的一个重要研究方向,这个问题在技术上依赖于如何根据已有检索返回结果对图片重新排序。本文针对现有的研究成果,提出了利用评测系统选择最优特征,以及基于全特征集合选取图像特征的RankBoost相似图片重排序改进算法,并开发实现了一个完整的实时相似图片重排序系统,并提出了基于信息损失率进行图像预分类的方法。在真实互联网图片检索环境下生成的数据集上做了大量的测试工作,并利用多种指标对图像重排序结果进行了评测。实验结果表明,利用信息损失率进行图片预分类RankBoost改进算法得到了优质的重排序结果,从而证明了预分类方法和改进算法的有效性。更多还原

【Abstract】 With the volume of digital images going up fabulously, people become to rely on computer to access the image more and more. Similar image retrieval, as an extension of general image retrieval, is a very important research field in medical applications, and is technically dependent on how to re-rank similar images returned by a keyword-based search. In this paper, we propose an improved RankBoost algorithm with selecting best features by evaluation system and feature selection on full feature set, and build a complete real time similar image re-ranking system. An image classification approach is proposed by the loss of information. We have done a lot of experiments in the data-sets generated in the real Internet image search environment. The result of image ranking is evaluated by several criteria. The experimental results on image database demonstrate that the proposed improved RankBoost algorithm has achieved excellent results in image re-ranking after image pre-classification by the loss of information approach, which confirms such a conclusion that the pre-classification and improved RankBoost algorithms are effective.更多还原