Improving the Efficiency of Information Retrieval in Digital Library Databases Using Image-Based Search

Abstract

The current research aims to improve information retrieval processes by enhancing the efficiency of search operations through image-based search. This is achieved by adding this feature to the search engines embedded in database systems. The researcher adopted an experimental approach by designing a program that allows searching through images stored in the central library databases of Al-Mustansiriyah University. A sample of digital files stored in the central library was taken, consisting of 50 PDF sources, to conduct search operations using image-based search. The research reached several key findings, the most important of which are:

• The conversion success rate reached 100% for all 50 sources.
• The techniques used for extracting digital features (Histogram, SIFT, SURF) demonstrated significant effectiveness in representing and analyzing image content. SIFT proved to be the best in identifying distinctive image points and recording them in the database, with a success rate of 96% in creating clear and precise descriptive points.
• Search time was reduced by 40% compared to retrieval operations based on ready-made software. The average search time for a single image was 1.2 seconds. The similarity criteria proved effective.
• The system demonstrated high accuracy in retrieving sources. Out of 50 query experiments, the correct source was retrieved 47 times, reflecting a success rate of 94%. The three errors were due to unclear images or those insufficiently representative of the source content.

The research also concluded with several recommendations, including:

• The necessity of developing the skills of university library staff to handle such databases, focusing on accurate data entry and ensuring the quality of images during the entry process.
• Enhancing the system to include textual search as an alternative or complement to image-based search, providing users with multiple options for accessing sources.