Paper Details

ABSTRACT

This concept of Automatic description generation for news is predicated on thesis of automatically generating captions for pictures that is very important for several image related applications. This model postulates that pictures and their textual descriptions are generated by a shared set of latent Variables (topics) and is trained on a weakly labeled dataset (which treats the captions and associated news articles as image labels). The theoretic surface realization model generates captions that are favorable to human generated captions. Given a news image I and its associated document D, produce a natural language caption that captures the pictures content given document. The training data so consists of document-image caption tuple during testing, we have a tendency to area unit given a document associated an associated image that we should generate a caption and therefore the cognitive content should contain 2 sorts of information, data concerning however the pictures (or image regions) corresponds to words and knowledge concerning however these words are often combined to form a humanreadable sentence.