Web-Based Image Search Refinement Via Feature Extraction And Ranking

Abstract

This thesis is concerned with the development of the Refined Image Search by Example (RISE) system, which carries out web-based image searches and attempts to refine the results by utilizing an image analysis and ranking algorithm. Initially, background research is carried out regarding existing frameworks and systems for web-based image retrieval. RISE is based on the assumption that the combination of a textual and an image query can yield more meaningful search results. Specifically, the image results of a text-based search are retrieved from the web using the Google Image Search JavaScript API. Then, a combination of texture, color, and brightness low-level features is extracted from the resulting images as well as from the query image using C#. These features were selected after investigating a larger set of possible features and feature combinations using Matlab. Lastly, a similarity measure is utilized in order to determine the ranking of each resulting image with respect to the query image. Thus, the RISE image search results are presented in a refined and more relevant order using the user interface that was developed in ASP.NET and C#.