Date of Award
Doctor of Philosophy (PhD)
Image search and retrieval based on content is very cumbersome task particularly when the image database is large. The accuracy of the retrieval as well as the processing speed are two important measures used for assessing and comparing the effectiveness of various systems.
Text retrieval is more mature and advanced than image content retrieval. In this dissertation, the focus is on converting image content into text tags that can be easily searched using standard search engines where the size and speed issues of the database have been already dealt with.
Therefore, image tagging becomes an essential tool for image retrieval from large image databases. Automation of image tagging has received considerable attention by many researchers in recent years. The optimal goal of image description is to automatically annotate images with tags that semantically represent the image content. The speed and accuracy of Image retrieval from large databases are few of the important domains that can benefit from automatic tagging.
In this work, several state of the art image classification and image tagging techniques are reviewed. We propose a new self-learning multilayered tagging framework that can address the limitations of current approaches and provide mutual accuracy improvement between the recognition layer and the annotation layer. Our results indicate that the proposed framework can improve the overall accuracy of information retrieval in a variety of image databases.
Masoud, Mohamed, "Enhancing Automatic Annotation for Optimal Image Retrieval." Dissertation, Georgia State University, 2018.