Tagging Image by Exploring Weighted Correlation between Visual Features and Tags

Automatic image tagging automatically label images with semantic tags, which significantly facilitate image search and organization. Existing tagging methods often derive the probabilistic or co-occurring tags from the visually similar images, which based on the image level similarity between images. It may result in many noisy tags due to the problem of semantic gap. In this paper, we propose a novel automatic tagging algorithm. It represents each test image with a bag of visual words and a measure to estimate the correlation between visual words and tags is designed. Then, for each test image, its visual words are weighted based on their importance, and the more important visual word contributes more to tag the test image. To tag a test image, we select the tags which have strong correlation with the greatly weighted visual words. We conduct extensive experiments on the real-world image dataset downloaded from Flickr. The results confirm the effectiveness of our algorithm.