Grant: Discovery Grants Program - Individual

Over the last few years our team has been developing the CINDI System. The main thrust of the Cindi system is cataloguing, in a systematic manner using a Semantic Header, the information resources accessible via the Internet or via a traditional dissemination channel. The digital library component of Cindi allows the provider of a resource to catalogue her document using an expert system to guide her in choosing correct subject terms. Subsequently, a searcher can locate the index entry and thence access the source document if it is on-line.  CINDI also implements a "pull mechanism" via a CindiRobot, which is guided by a filtering system to judge the relevance of sites for extending the digital library collection. Another component of CINDI is a Virtual Question Answering system which provides precise passages of relevant information from the collection. ConfSys is another pull component which has been used in the last four years to manage the administrative tasks of a number of conferences and add the documents (papers) to the CINDI collection. Challenges that needs to be resolved in this project is the matching of papers to reviewers while avoiding all possible onflict of interests. The improvement of the algorithms used in these components, their integration and tuning are tasks ahead of us.The image component of CINDI is concentrating on medical image databases and retrieval systems. Here we are investigating a content -based image retrieval framework of medical images. The framework under investigation consists of machine learning methods for image pre-filtering, similarity matching using statistical distance measures, and a relevance feedback scheme. We will investigate both supervised and unsupervised learning techniques to associate low-level global image features with their high level visual and semantic categories. We intend to incorporate better user perception subjectivity via a relevance feedback mechanism to dynamically update the query parameters and adjust the proposed matching functions.