Probabilistic Model for Summarizing Text and Automatically Annotating Images with Keywords

Annotation based image retrieval is more feasible to fulfill user requirements in a straight forward manner. Modern multi-media documents are not merely collections of words but can be a vast collection of related text, images and audio references. Images that do not coincide with textual data cannot be retrieved. Analyzing the pictures in large collections is a crucial problem. Search engines on the web retrieve images without viewing their content, simply by matching user queries against thematically collocated textual information which in turn limits the applicability. Methods are proposed to automatically generate captions for a picture from a weakly labeled data. So as to annotate images and generate captions a probabilistic suggestion with abstractive and extractive caption generation model prevails. Indeed, the abstractive model compares favorably to handwritten captions and is often superior to extractive methods. However the system is designed which is been used to realize the features of the images locally and is less grammatical. A phrase-based probabilistic model is framed to generate captions for images. To resolve such criteria images are experimented with global features in thematically co-located documents related to document structure such as titles, section of articles and also to exploit syntactic information more directly.