Discovering Semantic Relationships Among PDF Book Figures Using Contextual and Visual Similarity

摘要

Figures in books enhance the knowledge of users about facts, statistics, objects and concepts. Searching of figures and their associated textual information in the context of a specific topic from a collection of books needs plenty of time. Currently the solutions available in literature for figure searching are limited to scientific documents and are inapplicable to books due to their large sizes. These solutions are also domain dependent, uni-model, context ignorant and limited to their local repository due to which they give a large number of irrelevant search results and the figures are retrieved with limited information. Therefore, a generic, bi-model, dual repository based figure retrieval system is proposed to retrieve figures from books along with the contextual information to improve the understandability of the user. The system can establish semantic relationships between PDF book figures and can also relate figures to images on the web. The system can retrieve the figures with 91-96.5% precision and makes the navigation and searching of figures an effective and pleasant experience.