Computer Science Colloquium

With the advance of modern technology, multimedia data in various formats are becoming available at an explosive rate. Many emerging applications such as multimedia digital libraries, medical image databases and geographic information systems require accessing multimedia information based on their content. Content-based multimedia retrieval represents a promising technology to address this need. Traditional content-based multimedia retrieval techniques describe images/videos on the basis of low level features (such as color, texture, and shape) and support retrieval based on these features. However, retrieving images/videos via low-level features has been proven unsatisfactory, and new techniques are needed to support high-level queries. Research efforts are needed to bridge the gap between high-level semantics and low-level features.

My talk will introduce a novel approach to supporting semantics-based image retrieval. This approach is based on the concept of the monotonic tree, which consists of monotonic lines. An outward-falling/climbing monotonic line of a gray image is a boundary within the image which is characterized by higher/lower pixel values just inside the boundary than just outside. All monotonic lines in an image form a rooted tree, called a monotonic tree.

Monotonic tree is used as a hierarchical representation of image structures. The structural elements of an image are modeled as branches (or subtrees) of the monotonic tree of the image. These structural elements are classified and clustered on the basis of such properties as color, spatial location, harshness and shape. Each cluster corresponds to some semantic feature. Following these steps, images can be automatically annotated with category keywords, thus facilitating high-level (semantics-based) image querying and browsing. Monotonic tree has been applied to analyze scenery images, and locate image background. In future work, I intend to extend the monotonic tree model to general semantic categories on both images and videos.

Yuqing Song received his BS degree in Mathematics from Nanjing University in China in 1990, and his MS degree in Computer Science from Chinese Academy of Sciences in 1993. He expects to complete his PhD in Computer Science in State University of New York at Buffalo in May 2002. His research interests are multimedia, image processing, computer vision, and data miotion.