DR. SAMARABANDU is interested in the areas of computer vision and pattern recognition. He has been active in multi-dimensional image processing and image analysis frameworks during the past few years.
During his graduate studies and post-doctoral training, he was exploring various methodologies of image analysis systems and developed a system that incorporated both top-down and bottom-up methodologies. This served as the starting point for many of the analysis application that he later developed.
As a part of this research he developed many image analysis algorithms including the development of a texture characterization algorithm using morphological operations. This work is being continued today at the University at Buffalo where calcium loss in bones in zero gravity conditions is studied.
In order to effectively use these algorithms and to evaluate their accuracy and validity he developed a comprehensive image visualization environment for 3D images that incorporated both image data and graphical overlays for segmented structures. It allowed the user to query the morphometrical information about these segmented structures using a graphical interface. The system is still being used to analyze confocal images at the University of Buffalo.
These analysis and visualization tools made it possible to discover several important aspects of the dynamics of DNA replication sites in mammalian cells by a biology research group at the University at Buffalo.
Dr. Samarabandu is currently involved in developing a flexible framework for image analysis systems and proposes to address a crucial need in computer vision by encompassing expert knowledge within the system. This has the short term potential to provide a much higher level and more intuitive user interface for a complex image analysis system. Such an interface makes it readily available for non-expert users allowing widespread use in different domains. Currently almost all image analysis systems are either written specifically or configured by an expert for a given problem. Changing the problem domain would need expert help in re-configuring if at-all possible. In the long term, such a framework will help research machine learning algorithms and such advances would prove crucial in the next generation of intelligent systems.