KAZE算法在图像检索中的应用  

The application of KAZE algorithm in image retrieval

作　　者：卢建军[1] 李杰[1] 卫晨[2] 刘志鹏[2] 

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121 [2]西安邮电大学经济与管理学院,陕西西安710121

出　　处：《西安邮电大学学报》2017年第3期109-114,共6页Journal of Xi’an University of Posts and Telecommunications

摘　　要：给出一种KAZE算法与视觉词典库模型相结合的图像检索算法。利用KAZE算法提取特征点,构造特征描述子,进行预匹配;采用随机抽样一致性算法剔除误匹配点和错误点;通过K均值聚类算法对特征描述子进行聚类,建立视觉词典库模型。将图像特征描述子映射到视觉词典库模型,得到检索结果。实验结果表明,该算法可以检索相同或相似图像,图像检索的平均查准率为55%。An image retrieval algorithm combining KAZE algorithm with the visual dictionary model is proposed. In this algorithm, the KAZE algorithm is used to extract feature points, to construct feature descriptors, and to pre-match; the random sampling consistency algorithm is used to remove mismatched points and error points; and the K-Means algorithm is also used to characterize descriptors clustering. Then the visual dictionary model is established. The image feature descriptor is then mapped to the visual dictionary model, and the retrieval result is obtained. Experimental results show that the algorithm can retrieve the same or similar image, and the average precision of the image retrieval is 55%.

关 键 词：KAZE算法 随机抽样一致性算法 K均值聚类算法 词袋模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]