Research On The Retrieval Method Of Hot Topics Based On Image Indexing

With the rapid development of the Internet/The carriers of information is more and more diversified,and he dissemination of information is quicker and quicker.News, BBS, social networking sites and so on have became the main channels of the public to get information,But with the explosive increasing of the amount of information, through the text retrieval geting the interested topics has been unable to meet the demand.And images are the important carrier of information, which containe more information and more rich than words, and in the visual sense is more direct.In today’s Internet, image retrieval has been developing rapidly and has great application value, but the current image retrieval is simply to retrieve images by an image, howerver sometimes users have more interested in the topics hidded behind the images. Through the text we can retrieve the topics and information related the text, because of be not consistent to understand the information for different people,it leads that the key word retrieveed is different, and also it can appear differenct information or topic retrieved, however, image is objective existence relative to the text and has not subjective will, so how to retrieve the related topic by images is just what this article want to study.In this paper, through the research of the vast amounts of Internet data this paper proposes a fast method of retrievaling topics by image;Firstly when crawling data, we should combine of the images and text to crawl, making sure that each image or image set has a certain sum of relevant text.Secondly we can adopt the method of hierarchical clustering to find hot topics from text data crawled, and ensure high that hot topics have high polymerization.Then we must combine each topics with the high correlation images,and the images are maximum correlation sets of the topic. Finally we combine local sensitive hashing algorithm (LSH) and the weighted color correlogram to retrieve the topic related with images. It Can be found by experiments that this method can effectively retrieve the topics related with the query image,and both on the accuracy and speed it can meet a certain of requirements.