SEMANTIC SIGNATURE BASED IMAGE SEARCH IN WEB

Abstract

Web image retrieval may be a difficult task that needs efforts from image process, link structure analysis, and net text retrieval. Since content-based image retrieval continues to be thought-about terribly difficult, most current large-scale net image search engines exploit text and link structure to “understand” the content of the online pictures. However, lo- cal text info, like caption, filenames and adjacent text, isn't invariably reliable and informative. Therefore, international info ought to be taken under consideration once an internet image retrieval system makes connection judgment. During this paper, we have a tendency to propose a re-ranking methodology to enhance net image retrieval by rearrangement the pictures retrieved from a picture computer programme. The re-ranking method is predicated on a connection model that may be a probabilistic model that evalu-ates the connection of the hypertext markup language document linking to the image, and assigns a likelihood of connection. The experiment results showed that the re-ranked image retrieval achieved higher performance than original net image retrieval, suggesting the effectiveness of the re-ranking methodology.