Tech Reports

Tech Report kmi-08-08 Abstract

The Use of Ontologies for Improving Image Retrieval and Annotation
Techreport ID: kmi-08-08
Date: 2008
Author(s): Ainhoa Llorente Coto

Nowadays, digital photography is a common technology for capturing and archiving images due to the falling price of storage devices and the wide availability of digital cameras. Without efficient retrieval methods the search of images in large collections is becoming a painstaking work. Most of the traditional image search engines rely on keyword-based annotations because they lack the ability to examine image content. However, “a picture is worth a thousand words”, this means that up to a thousand words can be needed to describe the content depicted in a picture. This research proposes the use of highly structured annotations called ontologies to improve efficiency in image retrieval as well as to overcome the semantic gap that remains between user expectations and system retrieval capabilities. This work focuses on automated image annotation which is the process of creating a model that assigns visual terms to images because manual annotation is a time consuming and inefficient task. Up to now, most of the automated image annotation systems are based on a combination of image analysis and statistical machine learning techniques. The main objective of this research is to evaluate whether the underlying information contained in an ontology created from the vocabulary of terms used for the annotation could be effectively used together with the extracted visual information in order to produce more accurate annotations.