Research On WEB Image Automatic Annotation Method Based On Tag

With the development of network technology, the maturity of computer technology and the constantly growth of Internet user, internet produces large amounts of multimedia data every day while multimedia digital equipment and memory are constantly perfect and universal in hardware aspects. Image is an important part of multimedia data which contains rich semantic information, how to use the semantic information to effectively manage and use these Internet images is a problem we need to solve urgently. At present, At present, the label is a common representation of semantic information of Web image, this thesis mainly studies the Web image automatic annotation method based on tag,the details are as follows:First of all, Web image annotation is extracted by using image associated text with different positions, then use the Flickr API interfaces to return the label associated with the annotation words,according to the improved NGD algorithm to calculate the correlation between labels and get the final annotation of the image.Then, using the initial annotation to search the image in image community Flickr, the similar image sets is get from the image community Flickr by using k-nearest neighbor algorithm, the similarity between images is calculated by the underlying feature, the score of candidate image tags is calculated based on the image similarity weighted and keyword co-occurrence rate, select several number of the highest score tags as the final annotation of the image.Finally, the experimental results which based on this method show that the method proposed in the thesis can improve the performance of image annotation. At the same time,the method is simple and the method has practicability and value to a certain degree.