Image and Natural Language Processing for Multimedia Information Retrieval

Image annotation, the task of automatically generating description words for a picture, is a key component in various image search and retrieval applications. Creating image databases for model development is, however, costly and time consuming, since the keywords must be hand-coded and the process repeated for new collections. In this work we exploit the vast resource of images and documents available on the web for developing image annotation models without any human involvement. We describe a probabilistic framework based on the assumption that images and their co-occurring textual data are generated by mixtures of latent topics. Applications of this framework to image annotation and retrieval show performance gains over previously proposed approaches, despite the noisy nature of our dataset. We also discuss how the proposed model can be used for story picturing, i.e., to find images that appropriately illustrate a text and demonstrate its utility when interfaced with an image caption generator.