Md Mustafa Kamal Bhuiyan
Co-Supervisors: Dr. Orland Hoeber and Dr. Minglun Gong
Department of Computer Science
Wednesday, October 24, 2012, 12:00 p.m., Room EN 2022
It is challenging to design a method by which image search results are organized effectively and efficiently. Traditionally image search results are organized in paged grid layout where the results are ranked based on the relevance to the query. This is not feasible for large number of image search results due to the limited ability to explore and manipulate the search results. It is often tedious and time consuming process to browse sequentially. To overcome the problems of grid layout, many researchers have been using visual features of the images to organize the search results. This approach leads to the semantic gap, which is addressed by some other researchers. To avoid that problem, some researchers have been using only textual information associated with the images to organize the search results. However, this approach does not adequately address the problem of the semantic gap. In order to address these problems, this thesis is intended to establish a novel approach by organizing the image search results based on the combination of the visual and metadata features: user-generated tags, and device-generated temporal and spatial features. This will organize the images in such a way that the similar images will be placed close to each other. I am also proposing a dynamic weight adjustment feature by which the searchers will be able to reorganize their search results on the fly based on the different importance levels on the different features. An animated interface will be provided, which will represent the impact of dynamic changing of the features. This system will also provide a highly interactive search interface where the searchers will be able to explore and manipulate the search results effectively and efficiently. Finally, I will validate the resulting system through experiments on image search tasks.