|
Zhongfei (Mark) Zhang received B.S. (cum laude) in
Electronics Engineering, M.S. in Information Science,
both from Zhejiang University, Hangzhou, China, and
Ph.D. in Computer Science from the University of Massachusetts
at Amherst. He joined the faculty of Computer Science
Dept. at SUNY Binghamton in the Fall of 1999. He has
published over 70 peer-reviewed academic papers in
international and national journals and conferences
and several invited papers and book chapters, has edited
or co-edited two books, has served as reviewers or
program committee members for many international journals
and conferences, and has served as grant review panelists
for several governmental and private funding agencies.
He is a Senior Member of IEEE, a member of IEEE Computer
Society, a member of ACM, and a fellow of the Institute
for Student-Centered Learning at Binghamton University.
He is an Associate Editor for Pattern Recognition published
by Elsevier Science.
Research Interests
Under Dr. Zhang’s direction, the Computer Science Department’s Multimedia Research Lab conducts research in the areas of multimedia information indexing and retrieval, data mining and knowledge discovery, computer vision and image understanding, pattern recognition, bioinformatics, and machine learning. The University has built patent portfolios in a number of these disciplines, in particular in image annotation and retrieval, community discovery and generation, and object detection in image databases.
Image Annotation and Retrieval / Teaching Machine Learning
Efficient access to multimedia database content requires the ability to search and organize multimedia information. The most common static image retrieval systems use manually annotated description or metadata to look to match the image content. However, this manual annotation is very expensive. Furthermore, the annotation generated by manual labor is often rather limited in annotating the content of imagery. Binghamton University has developed an automatic image annotation and retrieval method based on a probabilistic semantic model in which the visual features and textual words are connected via a hidden layer to constitute the concepts to be discovered automatically to explicitly exploit the synergy between the two modalities.
See RB-209, "System and Method for Image Annotation and Multi-Modal Image Retrieval Using Probabilistic Semantic Models"
Data Mining / Discovering Communities of Complex and Massive Data Collections
This portfolio encompasses a framework to perform relational clustering analysis in a unified manner addressing various clustering tasks, including traditional attributes-based clustering, semi-supervised clusters, co-clustering, spectral-based clustering, and graph clustering. The model can identify cluster structures which we call communities consisting of not only each type of data objects but also the interaction patterns between different types of objects. The novel approach can make sense and use of multi-type relational data by clustering multi-types of objects simultaneously.
The over-arching advantage is that one can work with real-world instead of flat, feature-based data sets involving objects of multiple types that are related to each other. As an example within the Web search domain, the database consists of objects such as Web pages, search queries and the Web user; within the scientific publication domain, there are communities comprised of different types of objects such as papers, keywords, authors, and conferences. One algorithm enables multiple clusterings to be combined in support of a distributed data mining effort. The ability to combine is particularly useful when working with sensitive personal data where it is not possible to centrally collect information due to privacy protection safeguards.
The power of such tools revolves around the ability to reveal local and global groups and structures. One can apply various algorithms to discover ‘hidden’ connections in applications ranging from Web-based communities, shopping patterns, social networking, and task scheduling, from homeland security and anti-terrorism, to gene patterns in bioinformatics and to epidemiology studies.
See RB-239, "Soft Correspondence Ensemble Clustering" and RB-240 "Rational Clustering, Multi-Type Relational data, Collective Factorization On Related Matrices"
Aerial Imagery / Detecting Shadows and Other Time-Specific Objects in Image Databases
In the computer vision field, state of the art technology is challenged to automatically and accurately detect shadows and other objects in color aerial / overhead images. Binghamton scientists have generated an algorithm that works well with complex images and images taken under varied brightness and illumination conditions. This work has application beyond aerial surveillance analysis, for example to the construction of simulated (pilot) training environments, 3-D scientific reconstructions, and battlefield planning and operations, including automated target recognition and tracking.
See RB-176, "Hierarchical Static Shadow Detection For Color Aerial Images" and RB-142, "Fast Automatic Detectionof Independent Motion From Compressed Surveillance Video"
|
