Co-PI: Anil K. Jain, Department of CSE, Michigan State University
Recent years have witnessed the explosion of digitized images and videos. With the increasing popularity of digital cameras and video recorders, it has become much easier for people to acquire a large number of digital images and videos. Websites, such as Youtube and Flickr, allow users to upload their pictures and videos to share with others all around the world. Flickr hosted more than 4 billions photos in 2009 and receives, on average, over 60 million images per month. YouTube supports about 14.6 billion video views per month and over 150 million video clips that are searchable over Internet. The large size of online image/video databases, such as Flickr and YouTube, calls for efficient and robust algorithms for image/video retrieval. The object of this research project is to address the fundamental limitations of the Bag-of-Words (BoW) model by developing efficient methods for visual vocabulary construction and keypiont quantization that effectively explore the available side information about images. This project will also develop robust retrieval models that are resilient to the errors caused by keypoint quantization. The proposed components will be integrated into a fully functional system, and will be evaluated by experiments on large image databases with over 10 million images..