Image Annotation And Retrieval In A Digital Library

The explosive growth of multimedia data in large-scale information repositories such as digital library and Web poses new challenges to conventional information retrieval technologies. A typical digital library always features a huge amount of heterogeneous data and a variety of both expert and non-expert users far beyond controls of conventional information retrieval systems. Therefore, the ability to support semantic-level queries is becoming a necessity, which is however beyond the capability of the up-to-date multimedia retrieval techniques. We research and extend the conventional multimedia retrieval technologies in this work by proposing some key techniques and practical work based on CADAL project: the image analysis and retrieval techniques, automatic image annotation mechanism and running system developed for CADAL. These techniques and work provide critical building blocks for retrieval facilities such as CADAL or similar information repositories.We present an overview of the background, motivation, and basic approaches of our research in the beginning of this paper.In Chapter 2, we present a review of the research on image retrieval of different time periods and the brief introduction to automatic image annotation techniques.In Chapter3, we illustrate the necessity of developing automatic image annotation techniques, followed by the introduction to recent researches on automatic annotation techniques including the fundamental approaches and related works.In Chapter 4, we propose a novel algorithm for automatic image annotation. People have traditionally used manual annotation for linguistic indexing to support semantic-based image management and retrieval. However, this approach is liable to subjective, and requires a huge amount of human effort, especially for large image collections. Depends on annotated image collections, many researchers have proposed many automatic image annotation methods employ machine learning and statistical learning technologies. However most of them suffered from the skewed distribution of annotated keywords which greatly influenced the effectiveness of automatic image annotation. Our algorithm tried to alleviate such negative influence by considering the order information existed in the annotated image collections and the experiments show us positive results.In Chapter 5, we present the practical work we have done for CADAL project. We introduce the history of our image retrieval system, the background on which we developed it for CADAL and the functionalities and operations of this system, also wegive detailed description on constructing this compound-techniques system together with the problems we faced in different developing periods including designing, coding, testing and integrating. The experiences we gained by finding ways to solve such problems is presented in this chapter as well.Finally, we made a conclusion of whole paper in Chapter 6, with a brief discussion of the application prospect and future research directions.