Automatic Image Tagging
Seán Moran · 2009
Automatic Image Tagging seeks to assign relevant words (e.g. “jungle”, “boat”, “trees”) to images that describe the actual content found in the images without intermediate manual labelling. Current approaches are largely based on categorization, and treat the tags independently, so an annotation (jungle,trees) is just as plausible as (jungle,snow). The goal of this dissertation was to develop a probabilistic model (the Continuous Relevance Model) to take into account the dependencies between keywords so as to provide more precise annotations. The main findings suggest that, under certain conditions, taking into account keyword correlation, coupled with an efficient method (beam search) to search over sets of tags is an effective method to increase annotation accuracy.