Annotating images by semantic representation using the open knowledge base

Abstract

A good image annotation scheme is highly desired especially when images with crude description provide inadequate information and images with no description are not accessible by text based search. In the area of image annotation, this study aims to propose a new approach by combining image low level features and semantics available in open knowledge base. Image classification is one of the steps in image annotation. The best classifier was determined by conducting a comprehensive experiment where various machine learning algorithms performances were compared. Using feature extraction, initial tag population were generated by retrieving tags from the most similar images identified. Experiments were carried out to determine the best parameters that yield the best performance. Finally, tags related to domain of interest were given semantic meaning by optimizing ontologies and the open knowledge base. Comparing image annotation performance before and after linking to the open knowledge base is the main evaluation of this study. Evaluation is based on the standard performance metrics; precision, recall, and F-Measure. This study demonstrates that representing the identified concept of image annotation semantically is most useful in increasing image annotation performance.

Keywords: image annotation, image classification, semantic, open knowledge base