Efficient Phrase-Based Model for Automatic Caption Generation of Images

Abstract

This paper confront the trouble of automatic generation of caption for news images which are collocated with thematically  related  documents as well as the development of efficient tools that generate descriptions for images automatically which is more advantageous  to image search engines that get welfare from the image description in supporting more accurate and targeted queries for end users. This paper deals with the generation of captions from the database of images, news articles that captures the image's contents and consist of two factors that are content selection and surface realization. Content selection shows the relationship between the appearance of certain features in a document with the appearance of corresponding features in a given image whereas surface realization arbitrate verbalization of the chosen contents. The annotation process applies over the images and documents collection available in the database. To render the extracted image content in natural language without relying on rich knowledge resources, sentence-templates or grammars we need to be considering both extractive and abstractive caption generation models. In the paper we will scrutinize phrase based model for caption generation as a consequence with phrase tree construction. After making comparisons between obstructive and extractive method it is examining that output of attractive model is better than extractive method.