| With the advent of the era of information explosion, the data newly generated by the entire human community gets a spurt growth, which contains huge image data. In additional, the cost that people spend on the images and videos has been increasing with the popularity of the Internet. At the same time, the needs that people get something interested from the mass of image information are increasing too. Since 1970 s, the image retrieval technology has experienced the rapid development from text-based method to content-based method and the mixed mode of both. Given the limitations of the conventional image retrieval methods and the poor performance while the modern image retrieval conducts in complex background, this paper presents an image similarity retrieval technology based on the KAZE feature, and implements an image retrieval system based on Java Web.Firstly, this paper makes a brief introduction to the global feature in content based image retrieval technology, and then does a deeply research in the KAZE feature based on the local variables of image, and verifies the rationality and effectiveness by experiments using KAZE feature to build an image retrieval system. Subsequently, we introduce the feature matching algorithm in detail, and choose the Kd-Tree algorithm based on BBF as the feature vector matching algorithm for this paper. For the less mismatching points generated in the matching process, we use the RANSAC algorithm to solve the problem which improve the matching accuracy effectively. In particular, we build a visual dictionary database model to solve the retrieval inefficiencies problem in large scale of data and apply it to the image retrieval system in this paper. Finally, we use the J2 EE technologies building an image retrieval system based on Java Web.A series of experimental results in this paper show that the image retrieval system based on the proposed algorithm and the related solutions presents a better retrieval performance. |
PDF Full Text Request