Authors:
Sandro Aldo Aramini
;
Edoardo Ardizzone
and
Giuseppe Mazzola
Affiliation:
Università degli Studi di Palermo, Italy
Keyword(s):
Automatic Illustration, Story Picturing, Semantic Similarity, Semantic Space, Text Analysis, Web Image Search, Google Image.
Related
Ontology
Subjects/Areas/Topics:
Abstract Data Visualization
;
Computer Vision, Visualization and Computer Graphics
;
Text and Document Visualization
Abstract:
In this paper we propose a totally unsupervised and automatic illustration method, which aims to find onto
the Web a set of images to illustrate the content of an input short text. The text is modelled as a semantic
space and a set of relevant keywords is extracted. We compare and discuss different methods to create semantic
representations by keyword extraction. Keywords are used to query Google Image Search engine for
a list of relevant images. We also extract information from the Web pages that include the retrieved images,
to create an Image Semantic Space, which is compared to the Text Semantic Space in order to rank the list
of retrieved images. Tests showed that our method achieves very good results, which overcome those obtained
by using a state-of-the-art application. Furthermore we developed a Web tool to test our system and
evaluate results within the Internet community.