期刊名称：International Journal of Multimedia and Ubiquitous Engineering
印刷版ISSN：1975-0080
出版年度：2018
卷号：13
期号：1
DOI：10.21742/ijmue.2018.13.1.02
语种：English
出版社：SERSC
摘要：In this paper, a second-order retrieval algorithm and an improved bag-of-words algorithm were proposed and applied to search similar images. Features of an image were firstly extracted by the method of SIFT. The word frequent table of the features was then created by the improved bag-of-words algorithm which is a combination of hash algorithm and K- Means algorithm. Based on the word frequent table, similar images were finally identified by the second-order retrieval algorithm. However, the second-order retrieval algorithm includes two steps. The first-order retrieval is a cursory search and it pays much attention to the similarities between the distributions of features, the second-order retrieval is an accurate search and it depends on the proportion of the same feature points to the total points. The experiment results imply that this method has good performance on the aspect of recall, showing high accuracy and efficiency.
关键词：SIFT features;image retrieval;bag of word algorithm;feature word frequent table;second-order retrieval
