My research interests are around the following topics of information retrieval.
Video Information Retrieval:
Retrieval of information from video collections is an active research theme. One of the challenging problem is detecting shot and scene cuts in a video sequence. Currently, we are working towards detecting shots using macroblock information in compressed domain. In addition, we are investigating use of sound information for keyframe selection, scene analysis and video abstraction.
Image Information Retrieval:
We are investigating techniques for organising images semantically based on the pixel information.
It has been identified that combining evidence from multiple sources will improve the effectiveness of retrieval. I am interested in studying the application of the Dempster-Shafer theory for this purpose. After having simplified the computational complexity of the Dempster-Shafer evidence combination in the context of information retrieval, we are exploring its application in the context of retrieval from large collections such as those produced by the TREC.
Evaluation of IR systems:
The classical IR evaluation methodology based on measures like 'recall' and 'precision' fails in measuring various aspects in the context of multimedia and other interactive IR systems. As part of my Ph.D. investigation, I have developed a novel methodology for the evaluation of multimedia IR systems. It involves the use of real users in a simulated work task situation. As a continuation, we are iunvestigating issues related to the evaluation of various information retrieval systems.
Object-Oriented Software Development for IR:
During the last three decades research in the field of information retrieval has yielded a range of techniques for efficiently and effectively storing and retrieving information items based on their text content. However, many of such research techniques failed to reach practical deployment due to the difficulty and complexity involved in implementation. By employing the object-oriented software development paradigm we can build reusable software frameworks for IR and hence deliver advanced techniques to real users. We have developed a software framework called FLAIR. We are continuing this research by building reusable software frameworks and employing them for various applications.
Databases and Information Retrieval:
Modern IR applications demand management of complex documents which can range from being highly structured to highly unstructured. Researchers in IR and databases investigated the integration of the database and IR techniques. The hindrance in achieving this integration is the computational complexity of IR matching scheme. We approached this problem as an evidence combination problem and developed a retrieval model based on the Dempster-Shafer theory. We currently investigate the application of this technique for various semi-structured collections like world wide web and XML collections.
In addition to the above mentioned areas I have substantial interest in areas of digital libraries (3D image retrieval), and information management on the world wide web and retrieval of information using WAP enabled devices.