• home
• current students
• graduate
• events

Presenter:
Supervisor:

Date: Tue, March 22, 2005
Time: 14:30:00 - 15:30:00
Place: EOW 430

ABSTRACT

Abstract

In this presentation, I will give an overview of my past and current research on image processing and retrieval. Then I will give some possible directions for my future research.

My current research is about developing theoretical frameworks and designing algorithms for image and multimedia collection retrieval and indexing. Over the last four years, I made four main contributions. First, I performed a literature review on image retrieval and its application to the World Wide Web. Among the points I studied, there is the user s needs, applications of image retrieval, feature extraction, similarity measures, indexing, relevance feedback, feature selection, and query formulation. I studied existing techniques, their advantages and their limitations, the engines that adopted them, and made some propositions. This study allowed me to identify some issues that require an urgent intervention. In the subsequent chapters, I tried to resolve some of these issues. In the second chapter, I approached two main issues, namely, relevance feedback and feature selection. I proposed a model that, contrary to existing methods which use the positive example only, combines positive and negative examples in order to improve retrieval accuracy. I proved that this model allows better meeting the user s needs and understanding his/her judgement about the importance of each feature. In the third chapter, I proposed an improvement of the last work. Being based on a probabilistic model, this new mechanism allows, among other things, to better model the data, to interpret the obtained results and to perform a good feature selection. Furthermore, it enables to take into account multi-class queries and region-based queries. In the fourth chapter, I proposed a mechanism for image collection organisation, which I applied for indexing, browsing, retrieval, and summarization. The hierarchical index generated by this tool helps in reducing retrieval time and improving its accuracy. It also constitutes a browsing catalog that is very handy for users who do not have a clear idea in mind about what they are looking for. The fact that this work combines keywords with image content makes it possible to search images by their content as well as by their semantic concepts, which helps narrowing the semantic gap.

In addition to pursuing the development of solutions in my research domain, several questions have arisen from my investigations. They constitute promising directions for my future research. Future directions for my research include, but are not limited to:

• Application of probability theory in software development and validation.
• Studying the users of developed software in order to better meet their needs.
• Studying and proposing some solutions to the problem of high dimensions in data modeling.
• Developing a medical image database that is shared among hospital and health organisations, provided with its organisation tools.
• Cross media processing.
• Improving retrieval effectiveness by integrating Dirichlet mixtures (started recently).
• Studying similarity from a psychological point of view (started recently).
• Standardization in image and multimedia description.

Obviously, the course of action I pursue will be influenced by the collaborations and experiences afforded by my next position. Other research directions remain possible, especially in collaboration with my next colleagues. Broad fields in which I can act in the future include software engineering, probability theory and its applications, artificial intelligence, and signal processing.