Research And Implementation Of Tag Completing On Community Images Based On Semantic Context

In recent years, with the rapid development of Internet technology, the status of social media on the network has played a more and more important role. Meanwhile, influenced by the advancement of social media, some community websites which are based on images sharing are becoming increasingly active. As the size of the image increases and the number of image data grows, how to manage the image database and how to retrieve image has become urgent problems. Due to the different feelings and understanding of different users, the tags provided by users are often implicit and incomplete, these tags can not basically satisfy the need of precise description of the content of pictures. Most previous approaches are incapable of automatically revising the inaccurate tags and filling the incomplete tags. To address this issue, this paper proposes a image tag completion algorithm based on semantic context aiming to solve the problem of incomplete tags.Through reviewing and analyzing the existing research results, we have found that the existing methods only consider the completeness of tags, however, they all ignore the observed tag’s order. So this paper carries out a algorithm of image tag completion based on semantic context, and give the introduction of Pairwise Comparisons. Pairwise Comparisons is the method of making all elements to be compared together at the same time and getting the final result by sorting the data. Our method first build image-tag matrix and, through the method of paired comparison,optimize the weight parameters with which the scores of observed tags are greater than those unobserved ones. Then we take k-NN method to find the nearest picture to ensure similarity of image visual feature and use Google distance to define the semantic correlation of tags. After getting the complete image-tag matrix,we utilize tag’s weight score to sort tags and get the performance.This paper introduces the idea of Pairwise Comparisons, providing a new solution for the image-tag complement, which is important and significant. To test the effectiveness of the proposed algorithm in this paper, we have conducted a large number of experiments and analyses based on the ESP Game image dataset, IAPR TC12 image dataset and Corel 5K image dataset. The experimental results show that our algorithm can guarantee the order of observed tags, compared to the existing algorithm, and has great performance in accuracy above the state-of-the-art and the order of results consistent with user’s habit about tags.