Image Tag Refinement With Self Organizing Maps

Abstract

Nowadays, data sharing has become faster than ever. This speed demands novel search methods. Most popular way of accessing the data is to search its tag. Therefore, creating tags, captions from an image is a research area that gains reputation rapidly. In this study, we aim to refine image captions by utilizing Self Organizing Maps. We extract image and caption pairs as feature vectors and then cluster those vectors. Vectors with similar content clustered close to each other. With the help of those clusters, we hope to get some relevant tags that do not exist in the original tags. We performed extensive experiments and presented our initial results. According to these results, the proposed model performs reasonably well with a 54% precision score. Finally, we conclude our work by providing a list of future work. © 2019 IEEE.