Hybrid Information Retrieval Model For Web Images

TL;DR

This paper introduces a hybrid image retrieval system combining visual features and HTML metadata, including a novel term weighting scheme, achieving higher precision in web image retrieval.

Contribution

It presents a new hybrid model integrating graphical content and textual metadata, along with the VTF-IDF weighting scheme that leverages HTML structure for improved retrieval accuracy.

Findings

Achieved higher retrieval precision than existing models.

Demonstrated effectiveness of combining visual and textual features.

Validated the VTF-IDF scheme's advantage over traditional weighting methods.

Abstract

The Bing Bang of the Internet in the early 90's increased dramatically the number of images being distributed and shared over the web. As a result, image information retrieval systems were developed to index and retrieve image files spread over the Internet. Most of these systems are keyword-based which search for images based on their textual metadata; and thus, they are imprecise as it is vague to describe an image with a human language. Besides, there exist the content-based image retrieval systems which search for images based on their visual information. However, content-based type systems are still immature and not that effective as they suffer from low retrieval recall/precision rate. This paper proposes a new hybrid image information retrieval model for indexing and retrieving web images published in HTML documents. The distinguishing mark of the proposed model is that it is based on both graphical content and textual metadata. The graphical content is denoted by color features and color histogram of the image; while textual metadata are denoted by the terms that surround the image in the HTML document, more particularly, the terms that appear in the tags p, h1, and h2, in addition to the terms that appear in the image's alt attribute, filename, and class-label. Moreover, this paper presents a new term weighting scheme called VTF-IDF short for Variable Term Frequency-Inverse Document Frequency which unlike traditional schemes, it exploits the HTML tag structure and assigns an extra bonus weight for terms that appear within certain particular HTML tags that are correlated to the semantics of the image. Experiments conducted to evaluate the proposed IR model showed a high retrieval precision rate that outpaced other current models.