An Automatic Image Tagging Based on Word Co-Occurrence Analysis

Abstract

with the expansion of the Social Web and the digital cameras, storage capacities are widening with hundreds of photos shared through these applications. Most of the Social Web applications allow users to describe their photos by using tagging approach. However, since the tagging is an optional process, most of these photos were left untagged or with insufficient tags. Hence, it is difficult to search and retrieve these photos. Therefore, in order to overcome this issue, our research aims to develop an automatic tag propagation tool, which will enrich an initial tag with other related tags by using the tag recommendation based on the word co-occurrence analyses. This includes Dice, Cosine and Mutual Information. This analysis enables the tool to identify and suggest utilization of related tags based on Word similarity. Our evaluation shows that Dice and Cosine provide better tags candidate to recommendation as compared to Mutual Information. Therefore, we have combined the results from both analyses to be a candidate list to support the automatic tag propagation.