Title: Interactive Object Retrieval in Large Video Collections
Supervisor: Dr. Cathal Gurrin
Information retrieval technologies help people to maximize the value of information sources by enabling them to find information precisely and rapidly from massive archives. With the rapid growth of data capturing, storage, and transmission technologies, vast amounts of video data are publicly accessible for personal and enterprise customers and as a result, internet users spend more time on interaction with video content. Fast and efficient multimedia retrieval applications are widely needed for those archives. In this work, I analyse, index and retrieve videos based on the objects contained therein, not relying on Youtube-style manual annotation. A high quality and useful framework based on Machine Learning technologies is proposed for solving this computer vision problem. This research is being undertaken as part of the annual TRECVid video search comparative evaluation framework. In this talk, I will present the research methodologies, evaluation datasets and proposed experiments that form my PhD topic. It is expected that the research result can inform the next generation of multimedia information access tools which can make it easier for any user to locate content from ever increasing, unannotated, digital video archives.