In this dissertation, I investigate new approaches relevant to content-based image retrieval techniques. First, the MOD paradigm is proposed, a method for detecting salient points in images. These... Show moreIn this dissertation, I investigate new approaches relevant to content-based image retrieval techniques. First, the MOD paradigm is proposed, a method for detecting salient points in images. These salient points are specifically designed to enhance image retrieval accuracy by maximizing distinctiveness. Second, the multi-dimensional maximum likelihood similarity measure is presented, which removes a critical limitation in prior research in this area and provides an improved method of comparing image features. Third, a texture classification method based on low dimensional constructed texture features is introduced which have very low computational complexity and would be suitable for real time video understanding or interactive search of very large image databases. The new approaches are tested on well respected international test sets containing representative imagery. Show less
In my dissertation I investigate techniques for improving the state of the art in content-based image retrieval. To place my work into context, I highlight the current trends and challenges in my... Show moreIn my dissertation I investigate techniques for improving the state of the art in content-based image retrieval. To place my work into context, I highlight the current trends and challenges in my field by analyzing over 200 recent articles. Next, I propose a novel paradigm called __artificial imagination__, which gives the retrieval system the power to imagine and think along with the user in terms of what she is looking for. I then introduce a new user interface for visualizing and exploring image collections, empowering the user to navigate large collections based on her own needs and preferences, while simultaneously providing her with an accurate sense of what the database has to offer. In the later chapters I present work dealing with millions of images and focus in particular on high-performance techniques that minimize memory and computational use for both near-duplicate image detection and web search. Finally, I show early work on a scene completion-based image retrieval engine, which synthesizes realistic imagery that matches what the user has in mind. Show less
