|
With the the image application needs to exponential growth as well as images, video equipment , the popularity of the Internet , there are a growing number of image resources . How to effectively organize, manage , index these vast amounts of image data to user-friendly search is a huge challenge . Based retrieval of image annotation technology came into being , to some extent, to solve this problem. Image annotation techniques in sorting, classification, and plays an important role in organizing and retrieving Internet image . This paper proposed a web image annotation framework based on user interaction , the framework fully take into account the context of web 2.0 user interaction and multi-user feedback tagging system , and integration \the conventional tagging system such as human the subjective semantic gap solution to the problem . In this thesis, there are four aspects of work : First, put forward the integration of multi-mode image of a web 2.0 community site marked design , marking information in a variety of text forms used to collect in the process of human-computer interaction . Secondly , the proposed automatic annotation framework for target detection , the module will provide users with the automatic annotation for users to modify and provide feedback . After the results of user feedback will be database records and incorporated into a new training sample , in order to improve the labeling system performance . Proposed a simple and effective candidate keyword filtering algorithm , annotation can effectively reduce noise , to thereby obtain a better labeling results . Fourth, as a supplement , put forward a designated target site images and related text crawling reptiles technology , this centered the integrity and future work of the paper to make ready .
|